Next Article in Journal
A Network Analysis Approach to Detect and Differentiate Usher Syndrome Types Using miRNA Expression Profiles: A Pilot Study
Previous Article in Journal
Optimal Feature Selection and Classification for Parkinson’s Disease Using Deep Learning and Dynamic Bag of Features Optimization
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Review

Systematic Review of Deep Learning Techniques in Skin Cancer Detection

by
Carolina Magalhaes
1,2,
Joaquim Mendes
1,2,* and
Ricardo Vardasca
2,3
1
Faculdade de Engenharia, Universidade do Porto, 4200-465 Porto, Portugal
2
Instituto de Ciência e Inovação em Engenharia Mecânica e Engenharia Industrial, 4200-465 Porto, Portugal
3
ISLA Santarem, 2000-029 Santarem, Portugal
*
Author to whom correspondence should be addressed.
BioMedInformatics 2024, 4(4), 2251-2270; https://doi.org/10.3390/biomedinformatics4040121
Submission received: 9 October 2024 / Revised: 4 November 2024 / Accepted: 7 November 2024 / Published: 14 November 2024

Abstract

:
Skin cancer is a serious health condition, as it can locally evolve into disfiguring states or metastasize to different tissues. Early detection of this disease is critical because it increases the effectiveness of treatment, which contributes to improved patient prognosis and reduced healthcare costs. Visual assessment and histopathological examination are the gold standards for diagnosing these types of lesions. Nevertheless, these processes are strongly dependent on dermatologists’ experience, with excision advised only when cancer is suspected by a physician. Multiple approaches have surfed over the last few years, particularly those based on deep learning (DL) strategies, with the goal of assisting medical professionals in the diagnosis process and ultimately diminishing diagnostic uncertainty. This systematic review focused on the analysis of relevant studies based on DL applications for skin cancer diagnosis. The qualitative assessment included 164 records relevant to the topic. The AlexNet, ResNet-50, VGG-16, and GoogLeNet architectures are considered the top choices for obtaining the best classification results, and multiclassification approaches are the current trend. Public databases are considered key elements in this area and should be maintained and improved to facilitate scientific research.

1. Introduction

The skin is the largest organ of the human body [1]. Daily exposure to harmful elements, such as ultraviolet radiation, can result in neoplastic lesions that often progress to cancerous states [2]. There are two main types of skin cancer: melanoma and nonmelanoma. The former originates from melanocyte cells and is known for its aggressive growth and invasion rate, with early diagnosis being crucial for a good prognosis. The latter encompasses squamous cell carcinoma (SCC) and basal cell carcinoma (BCC). Despite its low tendency to spread to other tissues, its localized growth can cause major disfiguration and tissue destruction [3,4].
Visual examination of lesions by a medical physician is the first step for detection, and a biopsy can be advised to completely ascertain whether a lesion is cancerous or not, making the entire process painful, time-consuming, and dependent on physician proficiency [5,6]. However, sensitivity (SN) and specificity (SP) vary depending on the physicians experience, with experts presenting SN of 90%, but low SP (59%), mid-level while dermatologists and trainees present SN and SP of approximately 80% and 60% and 80% and 40%, respectively [7]. This can result in delays in accurate diagnosis.
To address this challenge, researchers have shifted their efforts toward the development of machine learning (ML) and deep learning (DL) approaches, in the hopes of facilitating early detection in a time-effective manner. While ML comprises multiple data-driven strategies, delivering outcomes based on detected patterns, DL employs artificial neural networks that mimic the structure and function of the human brain [8,9], thus proving remarkably effective in challenging pattern recognition tasks [10]. In the context of skin cancer diagnosis, these vast networks can be particularly useful, as extensive amounts of imaging data are generated and the identification of relevant features can be challenging for a human operator, especially in a timely manner [11,12]. Ultimately, they can reduce errors and subjectivity associated with the diagnostic process. The decrease in error rates through the use of DL applications not only aids clinicians in delivering more precise assessments, but also improves patient outcomes by reducing unnecessary biopsies and guaranteeing prompt responses for at-risk patients [13,14,15].
To thoroughly evaluate the efficiency of DL approaches for skin cancer diagnosis using image data, a systematic review was conducted. The paper is divided into five main sections. Section 2 defines the established research methodology, followed by study selection and qualitative analysis in Section 3. The results are discussed in Section 4, along with future prospects, and conclusions are presented in Section 5.

2. Methods

The present systematic review was performed to select and assess different approaches focused on skin cancer diagnosis using imaging data and deep neural networks. A portion of the collected data was assessed using statistical methods to summarize the results found. The Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) 2020 guidelines were used as the basis for the entire process [16].

2.1. Information Sources and Search Strategy

A careful research strategy was outlined before examining any databases. The bibliographic search was conducted for three reference sources, i.e., PubMed, IEEE Xplore, and Scopus. To guarantee the identification and extraction of relevant data, a set of keywords related to the main topic was selected. The logical operator ‘AND’ was applied in the search string formulation to find the maximum possible number of relevant reports. The search strings used were as follows: (deep learning [Title/Abstract]) AND (skin cancer [Title/Abstract]), (“All Metadata”: deep learning) AND (“All Metadata”: skin cancer) and (TITLE- ABS- KEY (deep learning) AND TITLE- ABS- KEY (skin cancer). The sources were consulted until June 2024. Duplicate removal was performed after the bibliographic search.

2.2. Synthesis Methods

The title and abstract of the found records were analyzed to exclude studies that did not fit within the scope of the conducted review. The remaining research papers were further evaluated based on the defined selection criteria. The first search criterion eliminated reports that focused on the use of DL techniques only for image analysis purposes, as the systematic review was focused on the implementation of DL strategies for decision-making processes. Thus, the second criterion eliminated studies that implemented classification stages not based on DL approaches. The third criterion excluded reports that covered skin disease classification and not skin tumors, as it could encompass skin pathologies of non-neoplastic nature, e.g., psoriasis and dermatitis, which is not the aim of this review. Articles centered solely on DL network theory were also excluded to guarantee that only reports with substantial contributions to clinical applications were assessed. Articles centered on the development of devices or equipment were disregarded in order to maintain the focus on methodology approaches and outcomes related to skin cancer classification. Studies written in languages other than English were also excluded to ensure that all included reports could be methodically examined by the research team. The seventh eligibility criterion excluded literature reviews, as they do not provide primary research data.

3. Results

3.1. Study Selection

Using the defined search strategy, 1978 records were identified as possible candidates during the bibliographic search. Of those, only 575 presented titles suitable for the current review’s theme. Abstract screening resulted in the exclusion of an additional 326 research papers, and the remaining 19 could not be retrieved. The application of the first eligibility criterion resulted in the exclusion of 18 records, while another 15 did not use DL approaches for classification. An additional 34 studies were eliminated for focusing on skin diseases other than skin neoplasms (n = 11), citing their work as studying DL network theories (n = 9), developing new equipment (n = 4), or being written in any language other than English (n = 3). In addition, review papers were not considered (n = 7).
In total, 163 records were subjected to qualitative assessment. The entire process is summarized in the PRISMA flow diagram included in Figure 1 [16].

3.2. Qualitative Synthesis

3.2.1. Drawbacks of Deep Learning

One of the main drawbacks of deep neural network (DNN) applications is the need for an extensive image dataset to ensure proper training. Harangi explored how this problem could be bypassed by the construction of an ensemble of DNNs [17]. The ensemble was composed of four different DNN architectures (GoogLeNet, AlexNet, ResNet, and VGGNet), and the results were obtained based on the weighted output of each network. Melanomas, nevi, and seborrheic keratosis (SK) were differentiated, and the average area under the curve (AUC) (0.891) exceeded each individual’s performance. A few months later, a similar aggregation was explored (GoogLeNet, AlexNet, and VGGNet) for the same purpose but with a slightly lower AUC (0.848) [18]. Another approach commonly used is transfer learning, where an architecture already trained in a given task is repurposed for a different classification job, as in [19,20,21]. In regard to the application of DL models for skin cancer classification, most models are pretrained with resourcing to ImageNet [22,23,24,25,26,27,28]. In contrast, Adjobo et al. implemented Gabor filters to extract spatial information from images and reduce the cost of feature extraction during the training stages of a CNN [29]. The proposed strategy yielded an accuracy (ACC) of 0.9639 for malignancy detection, surpassing the use of a simple network by 0.0237. The lack of sufficient image data also motivated Yao et al. to tackle small and imbalanced datasets [30]. A modified RandAugment augmentation strategy was implemented with regularization DropOut and DropBlock to address underrepresentation and overfitting prior to classification. The proposed method delivered high performance compared to other ensemble models. Pérez and Ventura also focused their research on this dual problem. Residual learning was implemented in the construction of a progressive generative adversarial network (GAN), where previous training blocks passed extra inputs to succeeding blocks. This process delivers artificially generated skin lesion images, largely increasing the available dataset [31]. When a simpler approach is needed to increase training instances, traditional methods of data augmentation, such as rotation, mirroring, color enhancement, and replication, have also been successfully implemented [32,33,34,35,36,37,38,39,40,41], as have other GAN-based technologies [42] or attention-cost-sensitive deep learning strategies [43].
Most DL-based approaches utilize complete images to perform feature learning, which can increase computational costs and complicate the identification of meaningful parameters. Wu et al. proposed a multi-input strategy using five blocks of high-resolution images as opposed to entire images, a down-sampled entire image (224 × 224), and a cropped portion of the leased area after enhancement via the CAM method [44], which reduced computational cost while preserving global context information (ACC = 0.884, SN = 0.767, SP = 0.963). The same group applied a CNN with attention learning to focus network attention on areas of the image that were important for the classification task, avoiding time loss due to unnecessary parameters (ACC = 0.857, AUC (ROC) = 0.837) [45]. A different strategy was used by Ashraf et al., who implemented a k-means algorithm to select a region of interest (ROI) prior to lesion classification [46]. The system outperformed the original whole-image approaches, delivering an ACC of 0.979 for melanoma and nevi differentiation. In a different approach, the technique of mask- and region-based CNNs was applied by Acosta et al. to select ROIs and then classify them with ResNet-125 (ACC = 0.904, SN = 0.82, SP = 0.925) [47]. Zhang et al. constructed a DL network with two branches: one focused on the extraction of local features (in four directions) and an additional one for global feature selection [48]. The complex feature extraction stage improved the ACC by 1.08% while keeping the size of the feature vector lower than that of more traditional approaches. Thus, the processing time was reduced while improving the performance. Another example is that of Osowski and Les, who presented an option based on grayscale value similarity that involved “flooding” only the leased area and surrounding tissue [49]. Segmented images were fed to an AlexNet-based CNN to create attributes to be used as inputs in a final classification stage composed of machine learning (ML) models (ACC = 0.886, SN = 0.661, SP = 0.896). Other examples of CNN applications for skin cancer detection with more extensive preprocessing stages can be found here [50,51,52,53,54,55,56,57,58,59,60,61,62]. Nevertheless, some authors see the use of whole raw images as an advantage due to the lack of complex and time-consuming segmentation and feature extraction steps [63,64].
Model interpretability is another challenge to be addressed, as deep learning strategies fail to present criteria to support the results encountered. This characteristic is probably the most crucial for the clinical adoption of a deep learning model, as clinicians need to comprehend the logic behind a model’s predictions. This insight shapes trust among physicians and patients, aids in model reliability validation, and encourages ethical implementation of AI in clinical decision-making. The local interpretable model-agnostic explanation (LIME) strategy was applied to a set of CNNs by Xiang and Wang to address this hurdle [65]. The VGG16, DenseNet, Xception, and InceptionResNet_v2 networks were combined to classify images of melanoma, nevus, and benign keratosis (HAM10000 dataset), and a GAN was applied during training for data augmentation. An ACC of 0.856 was achieved with LIME, highlighting lesion areas meaningful for the classification outcome. Tschandl et al. used content-based image retrieval to help the user understand the diagnosis made by a CNN network [66]. When a given image is loaded to the network, cases with similar anatomical characteristics are displayed to justify the network’s final diagnosis along with a prediction probability. The cases are selected based on deep feature similarity, with the approach suggested to be more helpful than softmax predictions in a clinical scenario. Nauta et al. developed a method to detect and quantify erroneous learning correlations and biases that occur during learning stages due to the presence/absence of given artifacts in specific types of skin tumor groups [67]. For instance, benign images in the ISIC dataset contained, e.g., elliptical colored patches that were not found in malignant cases. The authors artificially inserted these same artifacts in malignant cases of a test set and assessed the diagnostic abilities of a VGG16-based classifier previously trained on the biased dataset. A major increase in misdiagnoses was found, demonstrating the risks of applying black-box models in clinical settings when trained on possibly biased datasets. Different approaches for the implementation of explainable DL models can be found here [68,69,70,71,72].
Finally, DL models are usually quite extensive, hindering their implementation in devices with low computational power. To solve this problem, Maiti et al. compressed the DL structure through quantization-aware training [73]. This tactic allowed the differentiation of seven skin lesion types with a decrease of 6.5% in ACC compared to the original model, while reducing the processing time to one-quarter. An alternative approach to reducing the computational cost associated with DL training is discussed in [74].

3.2.2. Malignant vs. Benign

Identification of malignant tumors among benign tumors continues to be the primary goal in clinical practice. Thus, the use of novel DL strategies focused on this purpose is constant [75,76,77,78,79,80,81,82,83,84,85,86,87,88,89,90,91,92,93,94,95,96,97,98]. Albahar developed a regularizer technique based on the weight matrix’s standard deviation of the CNN to distinguish benign and malignant lesions [99]. This technique ensures that the weight matrix values remain similar, avoiding great dispersion. The greater the dispersion of the values is, the greater the penalty. The network was composed of layers: convolution (×2), pooling, dropout, flattening, dense-128 and output. The average ACC was 0.975 for different binary classification tasks (nevus vs. melanoma, SK vs. basal (BCC) and squamous cell carcinoma (SCC), SK vs. melanoma, and solar lentigo vs. melanoma). The lack of descriptions of the ACC, SN, and SP values for each individual task hindered comparison with other state-of-the-art methods. The author acknowledges the limitations of this technique, such as the difficulty in selecting the proper threshold to penalize the weight matrix due to the high time and computational costs. Afza et al. focused part of their work on perfecting the feature set used for classification [100]. The authors used a deep learning-based strategy for feature selection, combining a hybrid whale optimization and an entropy-mutual information approach to select the most suitable strategies, and then fused them with a modified canonical correlation-based method. This achieved an accuracy of 94.36% for multiclassification on the ISIC-2018 dataset. In contrast, Guo et al. supervised cross-entropy loss and covariance loss during CNN training to simultaneously rectify the extracted features and model outcome [101]. This approach yielded SN, SP, and ACC values of 0.942, 0.747, and 0.769, respectively. One of the larger test sets was used by Mijwil to distinguish between benign and malignant skin tumors [102]. Different CNN structures trained by knowledge transfer were applied, with InceptionV3 being the most suitable for the task (ACC = 0.869, SN = 0.861, SP = 0.876). In the future, the application of the model to a personal database is expected.
Identification of the type of malignancy, i.e., melanoma or nonmelanoma, is also essential, as many factors depend on the lesion type, e.g., prognosis and treatment course. Five different CNN structures were fine-tuned by Mishra et al. [102]. The networks were tested with and without dropout, as well as with and without the stochastic gradient descent (SCD) algorithm during training. An improved accuracy was found when both methods were used (ACC = 0.871, VGG-19 during training). Finally, Hagerty et al. separately applied dermoscopy features and patient clinical information as inputs to a DL network based on ResNet-50 [103]. The prediction score of each was then ensembled to reach an overall melanoma probability of 0.94.

3.2.3. Multiclassification

It is difficult to combine multiclass differentiation approaches with simpler machine learning approaches. The increased level of difficulty associated with the identification of distinguishable features for multiple skin tumor types is the main reason for this difficulty and has recently been addressed with regard to DL architectures [104,105,106,107,108,109,110,111,112,113,114,115,116,117,118,119,120,121,122,123,124,125,126,127]. Reisinho et al. sought to classify carcinomas, nevi, melanomas, and SK lesions with an ensemble of CNNs composed of ResNet, Inception_V3 and InceptionResNet_v2 [128]. The best output (ACC = 0.799, SP = 0.933, SN = 0.799) was achieved after passing the images through the three networks using their softmax output layers. The collected outputs were then concatenated and used as input, first to a dense layer (200 nodes with a 20% dropout), then to a 50-node dense layer, and finally to a softmax output layer. To address the multiclassification problem, Al-Masni individually tested the same networks with the addition of DenseNet 201 and differentiated seven skin lesion types in the International Skin Imaging Collaboration (ISIC) 2018 database. The authors fed segmented images to the networks, instead of whole-slide ones. The best overall result was found with ResNet-50 (ACC = 0.893, SP = 0.872, SN = 0.81) [129]. For the same dataset, Jasil and Ulagamuthalvi achieved an ACC of 0.77 with VGG16 [130] and DenseNet201 [131]. Heptaclassification was also performed by Sevli using a CNN structure to identify images from the HAM10000 dataset (ACC = 91.51%) [132]. This work was particularly relevant because the classification model was integrated into a web application and tested by seven dermatologists to assess its success in a clinical setting. The classifier achieved a correct classification of 90.28% of lesions and corrected 11.14% of the decisions made by the physicians. Similar work was performed by Rezvantalab et al., who compared the performance of different pretrained state-of-the-art networks with that of dermatologists [133]. Several DL models outperformed at least 11% of the physicians. Other reports of DL architectures outperforming those of dermatologists are easily found [134,135,136,137,138]. In contrast to performance comparisons, Hekler et al. chose to combine human and machine knowledge [139]. The expertise of 112 dermatologists and a CNN model achieved superior multiclassification results over individual performance.
Aburaed et al. balanced the HAM10000 dataset, obtaining approximately 1000 samples for each skin lesion type [140]. The VGG16 and VGG19 architectures and a deep CNN (with two convolutional layers and a max pooling layer repeated 3 times, followed by a flattening layer and two densely connected layers) were applied, with the latter being superior (ACC = 0.99, testing loss = 0.02). The validation images used were extracted from the training data and thus could have facilitated the classification task. For the same dataset, ElGhany et al. fine-tuned a ResNet50 model using regularization, batch normalization and hyperparameter optimization techniques, and a precision of 0.9609 was obtained [141]. An ACC of 0.947 was reported by Emara et al. with an Inception-V4 model [142]. The authors concatenated the features extracted from primary layers with those extracted from final layers to enhance the model’s performance. Gessert et al. constructed a more elaborate strategy to address the HAM10000 dataset [143]. Lesion diagnosis was guided by the knowledge used for ground truth annotation, meaning that if a given lesion was classified based on multiple parameters, e.g., expert opinion, biopsy, confocal microscopy, or more expensive methods, it was considered a lesion of difficult examination; thus, its loss of function was aggravated. Kondaveeti and Edupuganti used a similar strategy, but weights were adjusted for each class based on the frequency of the dataset (ACC = 0.90) [144]. Namozov et al. created a CNN architecture with a piecewise activation function, as opposed to more traditional options [145]. This alteration boosted network performance, even though it increased the computational cost. The seven different skin lesion types were differentiated with an ACC of 0.96 (only the training ACC was reported). Other reports of skin cancer classification with CNNs using the HAM10000 dataset can be found here [146,147,148,149,150,151,152].
Great results were achieved by Alkarakatly et al. for melanoma, nevus, and atypical nevus differentiation [153]. A CNN with five convolution layers, each containing a pooling layer, was constructed. The final layers presented a softmax activation function, while the remaining layers possessed a ReLU function. A total of 0.95, 0.94, and 0.97 were achieved for the ACC, SN, and SP, respectively. Hosny et al. distinguished melanoma and common and atypical nevi using a pretrained AlexNet (by replacing the last layer with a softmax layer with only three classes) [154]. The achievement of great results (ACC = 0.986, SN = 0.989, SP = 0.977) led the authors to successfully apply the same strategy to test the differentiation of melanoma from nevi or melanoma from nevi and SK (ACC = 0.977) [155]. Another uncommon classification task was solved by Serener and Serte, with an area under the curve (AUC) (ROC) of 0.80, where KC was detected using the Caffe framework and the ResNet-50, ResNet-18, and AlexNet architectures [156]. The same author subsequently presented a strategy for diagnosing melanoma and SK [157]. Lesion images were decomposed into seven directional sub-bands and used in conjunction as inputs for eight CNNs that worked in parallel to deliver eight probabilistic predictions. The fused outcome resulted in an average AUC (ROC) of 0.91 with the ResNest-18.
Iqbal et al. designed a deep CNN that performed eight-class classification using a small number of parameters and filters, despite its extensive number of layers and filter sizes [158]. With the ISIC-2019 dataset, the proposed approach delivered an ACC, SN, and SP of 0.90, 0.98, and 0.90, respectively. The authors propose the inclusion of additional clinical information, e.g., race, age, and sex, to further validate and improve the model. The work of Ahmed et al. surpassed these values in terms of the ACC, reaching 0.937 overall [159]. The same task was executed by Kassem et al. using a pretrained GoogLeNet structure [160]. Apart from delivering an ACC, SN, and SP of 0.949, 0.798, and 0.97, the model was also able to detect an unknown class of images. The authors created an index of similarity to identify a sample that did not belong to any of the classes described in the original dataset, mimicking a possible real-life medical scenario. Ahmed et al. chose a different strategy and treated instances of the unknown class as outliers, later applying one-class learning for the classification [161]. The results for this class were not very satisfactory.
To better address the challenges of class imbalance, Barata and Marques proposed a hierarchical diagnosis using DenseNet-161 [162]. The authors tested whether it was preferable to first classify lesions as malignant or benign, or to classify them as melanocytic or nonmelanocytic, as dermatologists do, and then proceed with the differentiation of classes. The former showed better results, but additional work with more classes of nonmelanocytic tumors is needed. Kaymak et al. followed this strategy to perform multiclassification, using a DL model to primarily differentiate melanocytic and nonmelanocytic lesions and a separate architecture to then detect malignant tumors among each type [163]. Malignant melanocytic lesions seemed to be diagnosed with greater accuracy (ACC = 0.84), while nonmelanocytic differentiation needed greater improvement. Finally, Moldovan implemented two-stage differentiation using a CNN to categorize lesions as nevus, melanoma, vascular lesions, or other types (ACC = 0.85) [164]. A second stage with the same model differentiated tumors included in the “other types” class as actinic keratosis (AK), BCC, benign keratosis, and dermatofibroma (ACC = 0.75). A greater number of training instances were indicated for future improvement. Han et al. took the challenge of multiclassification further with the development of a DL model that differentiated 12 different skin disease types, mostly skin tumors [165]. The network provided a possible list of predictions for a given lesion as a way to supply different impressions to a physician in clinical practice. The pretrained ResNet-152 was tested on two different datasets, delivering an average AUC, SN, and SP of 0.91, 0.864, and 0.855, respectively.
For multiclassification, some authors prefer the combination of multiple architectures to improve the performance and robustness of their classification model. Pratiwi et al. combined Inception_V3, Inception_ResNet_V2, and DenseNet_201 to differentiate seven skin lesion types [166]. After fine-tuning, the ensemble outperformed the individual networks, delivering an average accuracy of 0.972. Guo et al. trained several networks with an additive training approach [167]. This strategy boosts model predictions by repeatedly training the ensemble on incorrectly classified instances. Ultimately, an ACC of 0.852 was registered. For the same classification task, Shahin combined ResNet-50 and Inception_V3 to achieve an ACC = 0.899 without the use of a preprocessing stage [168]. A classification strategy based on the same model, Inception_V3, was used to determine the top ACC after optimization with Bayesian tuning of the training hyperparameters for multiclassification of pigmented skin lesions [169]. Benign and malignant pigmented skin lesions were differentiated by an ACC of 0.964 and an AUC of 0.98. The authors suggest the use of a preprocessing stage as an improvement for future tests, with the goal of increasing the performance metrics achieved with raw images. Other recent approaches focused on parameter tuning and optimization can be found in [170,171,172]. To ease multiclassification tasks, several authors have tested the possibility of incorporating patient information in the decision-making process [173,174,175]. Pacheco and Krohling presented an algorithm that enhanced feature maps extracted during classification through the incorporation of metadata [176]. This strategy made the model more aware of important features, improving its performance metrics for multiclassification (ACC = 0.91). Ou et al. also showed that the incorporation of patient metadata can significantly enhance a model’s performance [177]. The combination of information retrieved from images and meta features using an approach based on intramodality self-attention and intermodality cross-attention delivered an ACC of 0.768 and an ROC of 0.947.
The data retrieved from each record are summarized in Table 1 and Table 2.

4. Discussion

The qualitative synthesis results showed a clear shift over the years regarding classification tasks (Table 1 and Table 2). In the first deep learning applications for skin cancer diagnosis, the authors focused their research on mostly binary classification, particularly for melanoma identification, often with a smaller dataset and a single network. The evolution of multiclassification occurred naturally and was accompanied by experiments with larger datasets and ensemble models to take full advantage of the abilities of each architecture to solve more challenging problems. Some authors even choose to perform a two-stage classification phase, first differentiating melanocytic from nonmelanocytic tumors or benign from malignant lesions, and then proceeding with the identification of the remaining types [162,163,164]. This approach seems to yield better results than overall lesion type recognition. The use of preprocessing strategies to boost network classification outcomes is a controversial topic, with some authors considering this approach a valuable stage [50,51,52,53,54,56,57,62] and others ignoring it and considering that its possible benefits do not outweigh the larger amounts of time spent with code development and image processing [64,65]. The debatable use of preprocessing stages brings awareness to the importance of standardizing methodologies within the field, as variability could result in different classification outcomes, thus hindering comparison of findings among studies.
The vast majority of authors choose to employ a pretrained network, such as transfer learning, saving time in the conception and training of a full CNN, e.g., [133,154,155,160,165]. The favored architectures for classification include AlexNet and ResNet-50, while VGG-16 and GoogLeNet are also recognized as valuable models. Thus, only a small number of studies have verified a clear benefit in the conception and training of entire networks. (Table 1 and Table 2). The training/testing split suggested by most researchers indicates that a 70/30% or 80/20% ratio ensures adequate training while delivering the best results (Table 1 and Table 2). Recent studies have tended to employ an additional set for validation prior to testing the model on unseen data, e.g., [112,113,117,118,151]. Not all the authors employ a validation set in their research. Finally, there is a clear tendency to employ images from public databases, e.g., ISIC 2016, 2017, 2018, 2019, PH2, HAM10000 and MED-DONE (Table 1 and Table 2). The use of openly available images facilitates research by saving time involved in obtaining ethical approval from hospitals, performing image collection and network training, and providing monetary resources involved in material acquisition. The evaluation of the attained results with those of other authors was also facilitated by these databases. The authors can easily compare the applied methodologies to the same database and identify possible errors or weaknesses to further improve the achieved outcomes or even prove that the proposed approach surpasses previous ones. However, this comparison was not found in all the papers, which is a limitation of some related works, e.g., [131,143,165,167]. Other limitations include a deficient report of performance metrics, e.g., [22,25,73,99,103], hampering even more comparisons between records and the execution of a more complete quantitative analysis. The lack of description of training and testing conditions is also a fault in some records, e.g., [75,77,136], as these stages are crucial for general interpretation of the success of a given approach.
Future improvements suggested by some of the authors indicate the need for larger and richer datasets as a means of enhancing classifiers. The importance of a balanced dataset with a distribution of classes is emphasized to decrease misdiagnosis [18,65,102,109,130,132,133,134,135,146,162,164,165]. The implementation of GANs is presented as a possible solution, as are different techniques for data augmentation [32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65]. The use of a hierarchical diagnosis also appears to be beneficial [162]. The inclusion of patient clinical information, such as age, sex, and eye color, in the classification network is recommended by many, as there is clear growth in the application of metadata as additional information for the proposed classification model. This type of information is usually available during a real-life clinical diagnosis, providing the same amount of data to the classifier that is available for the physician [103,133,140,158,173,174,176,177]. The study of more preprocessing options is also recommended as an alternative to possibly upgrade classification metrics [33,42,81,109,164]. Many authors plan to improve the network architecture to detect additional skin tumor types and even recognize those that were not included during training stages, thus functioning as a real-life situation [22,35,109,141,161,162]. Finally, the ultimate goal is the implementation of the constructed models in a clinical setting both to identify possible weaknesses and facilitate their integration and acceptance by medical professionals [23,73,86,106,132,138,147].

5. Conclusions

This review focused on the assessment of deep learning applications for skin cancer detection through image data. A growing tendency for this topic over the last year was found, with classification tasks becoming harder and DL models becoming more complex. Models based on the AlexNet, ResNet-50, VGG-16 and GoogLeNet architectures are indicated as leading choices for top results. Multiclassification seems to be the current trend, with hierarchical approaches delivering better results. Public databases are key aspects of research concerning skin cancer detection and should be improved and enriched as a way of facilitating further developments in this field of study.

Author Contributions

C.M. and R.V. conceptualized and designed this review and performed the literature search. C.M. drafted the manuscript. R.V. and J.M. revised and edited the drafted manuscript. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by Project LAETA [grant numbers UIDB/50022/2020, UIDP/50022/2020] and the PhD Scholarship supported by FCT (national funds through Ministério da Ciência, Tecnologia e Ensino Superior (MCTES)) and co-funded by ESF through the Programa Operacional Regional do Norte (NORTE 2020) (EU funds)) [grant number SFRH/BD/144906/2019].

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

Not applicable.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Hunter, J.; Savin, J.; Dahl, M. The function and structure of the skin. In Clinical Dermatology, 3rd ed.; Blackwell Science: Hoboken, NJ, USA, 2002; pp. 7–28. [Google Scholar]
  2. Hunter, J.; Savin, J.; Dahl, M. Skin Tumours. In Clinical Dermatology, 3rd ed.; Blackwell Science: Hoboken, NJ, USA, 2002; pp. 253–282. [Google Scholar]
  3. Gruber, P.; Zito, P.M. Skin Cancer; StatPearls: Treasure Island, FL, USA, 2020. [Google Scholar]
  4. Ricotti, C.; Bouzari, N.; Agadi, A.; Cockerell, C.J. Malignant Skin Neoplasms. Med. Clin. N. Am. 2009, 93, 1241–1264. [Google Scholar] [CrossRef] [PubMed]
  5. Fink, C.; Haenssle, H.A. Non-invasive tools for the diagnosis of cutaneous melanoma. Ski. Res. Technol. 2017, 23, 261–271. [Google Scholar] [CrossRef] [PubMed]
  6. Piccolo, D.; Ferrari, A.; Peris, K.; Daidone, R.; Ruggeri, B.; Chimenti, S. Dermoscopic diagnosis by a trained clinician vs. a clinician with minimal dermoscopy training vs. computer-aided diagnosis of 341 pigmented skin lesions: A comparative study. Br. J. Dermatol. 2002, 147, 481–486. [Google Scholar] [CrossRef] [PubMed]
  7. Przystalski, K.; Leszek, N.; Ogorzałek, M.; Surówka, G. Decision Support System for Skin Cancer Diagnosis. In Proceedings of the Ninth International Symposium on Operations Research and Its Applications (ISORA’10), Chengdu, China, 19–23 August 2010; pp. 406–413. [Google Scholar]
  8. Kononenko, I. Machine learning for medical diagnosis: History, state of the art and perspective. Artif. Intell. Med. 2001, 23, 89–109. [Google Scholar] [CrossRef] [PubMed]
  9. Razmjooy, N.; Ashourian, M.; Karimifard, M.; Estrela, V.V.; Loschi, H.J.; Nascimento, D.D.; França, R.P.; Vishnevski, M. Computer-aided Diagnosis of Skin Cancer: A Review. Curr. Med. Imaging Former. Curr. Med. Imaging Rev. 2020, 16, 781–793. [Google Scholar] [CrossRef]
  10. Bakator, M.; Radosav, D. Deep Learning and Medical Diagnosis: A Review of Literature. Multimodal Technol. Interact. 2018, 2, 47. [Google Scholar] [CrossRef]
  11. Lee, J.-G.; Jun, S.; Cho, Y.-W.; Lee, H.; Kim, G.B.; Seo, J.B.; Kim, N. Deep Learning in Medical Imaging: General Overview. Korean J. Radiol. 2017, 18, 570–584. [Google Scholar] [CrossRef]
  12. Razzak, M.I.; Naz, S.; Zaib, A. Deep Learning for Medical Image Processing: Overview, Challenges and the Future. arXiv 2017, arXiv:1704.06825v1. [Google Scholar]
  13. McNeil, B.J.; Keeler, E.; Adelstein, S.J. Primer on Certain Elements of Medical Decision Making. N. Engl. J. Med. 1975, 293, 211–215. [Google Scholar] [CrossRef]
  14. Wijesinghe, I.; Gamage, C.; Perera, I.; Chitraranjan, C. A Smart Telemedicine System with Deep Learning to Manage Diabetic Retinopathy and Foot Ulcers. In Proceedings of the 2019 Moratuwa Engineering Research Conference (MERCon), Moratuwa, Sri Lanka, 3–5 July 2019; pp. 686–691. [Google Scholar]
  15. Shihadeh, J.; Ansari, A.; Ozunfunmi, T. Deep Learning Based Image Classification for Remote Medical Diagnosis. In Proceedings of the 2018 IEEE Global Humanitarian Technology Conference (GHTC), San Jose, CA, USA, 18–21 October 2018; pp. 1–8. [Google Scholar]
  16. Page, M.J.; McKenzie, J.E.; Bossuyt, P.M.; Boutron, I.; Hoffmann, T.C.; Mulrow, C.D.; Shamseer, L.; Tetzlaff, J.M.; Akl, E.A.; Brennan, S.E.; et al. he PRISMA 2020 statement: An updated guideline for reporting systematic reviews. BMJ 2021, 372, 71. [Google Scholar] [CrossRef]
  17. Harangi, B. Skin lesion classification with ensembles of deep convolutional neural networks. J. Biomed. Inform. 2018, 86, 25–32. [Google Scholar] [CrossRef] [PubMed]
  18. Harangi, B.; Baran, A.; Hajdu, A. Classification of Skin Lesions Using an Ensemble Of Deep Neural Networks. In Proceedings of the 2018 40th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), Honolulu, HI, USA, 17–21 July 2018; pp. 2575–2578. [Google Scholar]
  19. Anand, V.; Gupta, S.; Altameem, A.; Nayak, S.R.; Poonia, R.C.; Saudagar, A.K.J. An Enhanced Transfer Learning Based Classification for Diagnosis of Skin Cancer. Diagnostics 2022, 12, 1628. [Google Scholar] [CrossRef] [PubMed]
  20. Ghazal, T.M.; Hussain, S.; Khan, M.F.; Said, R.A.T.; Ahmad, M. Detection of Benign and Malignant Tumors in Skin Empowered with Transfer Learning. Comput. Intell. Neurosci. 2022, 2022, 4826892. [Google Scholar] [CrossRef] [PubMed]
  21. Abdelhafeez, A.; Mohamed, H.K.; Maher, A.; Khalil, N.A. A novel approach toward skin cancer classification through fused deep features and neutrosophic environment. Front. Public Health 2023, 11, 1123581. [Google Scholar] [CrossRef] [PubMed]
  22. Jiahao, W.; Xingguang, J.; Yuan, W.; Luo, Z.; Yu, Z. Deep Neural Network for Melanoma Classification in Dermoscopic Images. In Proceedings of the 2021 IEEE International Conference on Consumer Electronics and Computer Engineering (ICCECE), Guangzhou, China, 15–17 January 2021; pp. 666–669. [Google Scholar]
  23. Jibhakate, A.; Parnerkar, P.; Mondal, S.; Bharambe, V.; Mantri, S. Skin lesion classification using deep learning and image processing. In Proceedings of the 2020 3rd International Conference on Intelligent Sustainable Systems (ICISS), Palladam, India, 3–5 December; pp. 333–340.
  24. Manzo, M.; Pellino, S. Bucket of Deep Transfer Learning Features and Classification Models for Melanoma Detection. J. Imaging 2020, 6, 129. [Google Scholar] [CrossRef]
  25. Zhang, Y.; Wang, C. SIIM-ISIC Melanoma Classification with DenseNet. In Proceedings of the 2021 IEEE 2nd International Conference on Big Data, Artificial Intelligence and Internet of Things Engineering (ICBAIE), Nanchang, China, 26–28 March 2021; pp. 14–17. [Google Scholar]
  26. Zunair, H.; Ben Hamza, A. Melanoma detection using adversarial training and deep transfer learning. Phys. Med. Biol. 2020, 65, 135005. [Google Scholar] [CrossRef]
  27. Elaziz, M.A.; Dahou, A.; Mabrouk, A.; El-Sappagh, S.; Aseeri, A.O. An Efficient Artificial Rabbits Optimization Based on Mutation Strategy For Skin Cancer Prediction. Comput. Biol. Med. 2023, 163, 107154. [Google Scholar] [CrossRef]
  28. Ogundokun, R.O.; Li, A.; Babatunde, R.S.; Umezuruike, C.; Sadiku, P.O.; Abdulahi, A.T.; Babatunde, A.N. Enhancing Skin Cancer Detection and Classification in Dermoscopic Images through Concatenated MobileNetV2 and Xception Models. Bioengineering 2023, 10, 979. [Google Scholar] [CrossRef]
  29. Adjobo, E.C.; Mahama, A.T.S.; Gouton, P.; Tossa, J. Proposition of convolutional neural network based system for skin cancer detection. In Proceedings of the 2019 15th International Conference on Signal-Image Technology & Internet-Based Systems (SITIS), Sorrento, Italy, 26–29 November 2019; pp. 35–39. [Google Scholar]
  30. Yao, P.; Shen, S.; Xu, M.; Liu, P.; Zhang, F.; Xing, J.; Shao, P.; Kaffenberger, B.; Xu, R.X. Single Model Deep Learning on Imbalanced Small Datasets for Skin Lesion Classification. IEEE Trans. Med. Imaging 2022, 41, 1242–1254. [Google Scholar] [CrossRef]
  31. Pérez, E.; Ventura, S. Progressive growing of Generative Adversarial Networks for improving data augmentation and skin cancer diagnosis. Artif. Intell. Med. 2023, 141, 102556. [Google Scholar] [CrossRef]
  32. Mohamed, E.H.; El-Behaidy, W.H. Enhanced Skin Lesions Classification Using Deep Convolutional Networks. In Proceedings of the 2019 Ninth International Conference on Intelligent Computing and Information Systems (ICICIS), Cairo, Egypt, 8–10 December 2019; pp. 180–188. [Google Scholar]
  33. Muresan, H.-B. Skin Lesion Diagnosis Using Deep Learning. In Proceedings of the 2019 IEEE 15th International Conference on Intelligent Computer Communication and Processing (ICCP), Cluj-Napoca, Romania, 5–7 September 20119; pp. 499–506. [Google Scholar]
  34. Rezaoana, N.; Hossain, M.S.; Andersson, K. Detection and Classification of Skin Cancer by Using a Parallel CNN Model. In Proceedings of the 2020 IEEE International Women in Engineering (WIE) Conference on Electrical and Computer Engineering (WIECON-ECE), Bhubaneswar, India, 26–27 December 2020; pp. 380–386. [Google Scholar]
  35. Hassan, S.R.; Afroge, S.; Mizan, M.B. Skin Lesion Classification Using Densely Connected Convolutional Network. In Proceedings of the 2020 IEEE Region 10 Symposium (TENSYMP), Dhaka, Bangladesh, 5–7 June 2020; pp. 750–753. [Google Scholar]
  36. Thao, L.T.; Quang, N.H. Automatic skin lesion analysis towards melanoma detection. In Proceedings of the 2017 21st Asia Pacific Symposium on Intelligent and Evolutionary Systems (IES), Hanoi, Vietnam, 15–17 November 2017; pp. 106–111. [Google Scholar]
  37. Jain, S.; Singhania, U.; Tripathy, B.; Nasr, E.A.; Aboudaif, M.K.; Kamrani, A.K. Deep Learning-Based Transfer Learning for Classification of Skin Cancer. Sensors 2021, 21, 8142. [Google Scholar] [CrossRef] [PubMed]
  38. Alam, T.M.; Shaukat, K.; Khan, W.A.; Hameed, I.A.; Almuqren, L.A.; Raza, M.A.; Aslam, M.; Luo, S. An Efficient Deep Learning-Based Skin Cancer Classifier for an Imbalanced Dataset. Diagnostics 2022, 12, 2115. [Google Scholar] [CrossRef] [PubMed]
  39. Montaha, S.; Azam, S.; Rafid, A.K.M.R.H.; Islam, S.; Ghosh, P.; Jonkman, M. A shallow deep learning approach to classify skin cancer using down-scaling method to minimize time and space complexity. PLoS ONE 2022, 17, e0269826. [Google Scholar] [CrossRef] [PubMed]
  40. Abbas, Q.; Daadaa, Y.; Rashid, U.; Ibrahim, M.E.A. Assist-Dermo: A Lightweight Separable Vision Transformer Model for Multiclass Skin Lesion Classification. Diagnostics 2023, 13, 2531. [Google Scholar] [CrossRef] [PubMed]
  41. Azeem, M.; Kiani, K.; Mansouri, T.; Topping, N. SkinLesNet: Classification of Skin Lesions and Detection of Melanoma Cancer Using a Novel Multi-Layer Deep Convolutional Neural Network. Cancers 2023, 16, 108. [Google Scholar] [CrossRef]
  42. Qin, Z.; Liu, Z.; Zhu, P.; Xue, Y. A GAN-based image synthesis method for skin lesion classification. Comput. Methods Programs Biomed. 2020, 195, 105568. [Google Scholar] [CrossRef]
  43. Ravi, V. Attention Cost-Sensitive Deep Learning-Based Approach for Skin Cancer Detection and Classification. Cancers 2022, 14, 5872. [Google Scholar] [CrossRef]
  44. Wu, J.; Hu, W.; Wang, Y.; Wen, Y. A Multi-Input CNNs with Attention for Skin Lesion Classification. In Proceedings of the 2020 IEEE International Conference on Smart Cloud (SmartCloud), Washington, DC, USA, 6–8 November 2020; pp. 78–83. [Google Scholar]
  45. Wu, J.; Hu, W.; Wen, Y.; Tu, W.; Liu, X. Skin lesion classification using densely connected convolutional networks with attention residual learning. Sensors 2020, 20, 7080. [Google Scholar] [CrossRef]
  46. Ashraf, R.; Afzal, S.; Rehman, A.U.; Gul, S.; Baber, J.; Bakhtyar, M.; Mehmood, I.; Song, O.-Y.; Maqsood, M. Region-of-Interest Based Transfer Learning Assisted Framework for Skin Cancer Detection. IEEE Access 2020, 8, 147858–147871. [Google Scholar] [CrossRef]
  47. Jojoa Acosta, M.F.; Caballero Tovar, L.Y.; Garcia-Zapirain, M.B.; Percybrooks, W.S. Melanoma diagnosis using deep learning techniques on dermatoscopic images. BMC Med. Imaging 2021, 21, 6. [Google Scholar] [CrossRef]
  48. Zhang, D.; Li, A.; Wu, W.; Yu, L.; Kang, X.; Huo, X. CR-Conformer: A fusion network for clinical skin lesion classification. Med. Biol. Eng. Comput. 2023, 62, 85–94. [Google Scholar] [CrossRef] [PubMed]
  49. Osowski, S.; Les, T. Deep Learning Ensemble for Melanoma Recognition. In Proceedings of the 2020 International Joint Conference on Neural Networks (IJCNN), Glasgow, UK, 19–24 July 2020; pp. 1–7. [Google Scholar]
  50. Nasr-Esfahani, E.; Samavi, S.; Karimi, N.; Soroushmehr, S.; Jafari, M.; Ward, K.; Najarian, K. Melanoma detection by analysis of clinical images using convolutional neural network. In Proceedings of the 2016 38th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), Orlando, FL, USA, 16–20 August 2016; Volume 2016, pp. 1373–1376. [Google Scholar]
  51. Naronglerdrit, P.; Mporas, I.; Paraskevas, M.; Kapoulas, V. Melanoma Detection from Dermatoscopic Images using Deep Convolutional Neural Networks. In Proceedings of the 2020 International Conference on Biomedical Innovations and Applications (BIA), Varna, Bulgaria, 24–27 September 2020; pp. 13–16. [Google Scholar]
  52. Bhimavarapu, U.; Battineni, G. Skin Lesion Analysis for Melanoma Detection Using the Novel Deep Learning Model Fuzzy GC-SCNN. Healthcare 2022, 10, 962. [Google Scholar] [CrossRef] [PubMed]
  53. Lu, X.; Zadeh, Y.A.F.A. Deep Learning-Based Classification for Melanoma Detection Using XceptionNet. J. Healthc. Eng. 2022, 2022, 2196096. [Google Scholar] [CrossRef] [PubMed]
  54. Behara, K.; Bhero, E.; Agee, J.T. Skin Lesion Synthesis and Classification Using an Improved DCGAN Classifier. Diagnostics 2023, 13, 2635. [Google Scholar] [CrossRef]
  55. Putten, E.V.; Kambod, A.; Kambod, M. Deep residual neural networks for automated Basal Cell Carcinoma detection. In Proceedings of the 2018 IEEE EMBS International Conference on Biomedical & Health Informatics (BHI), Las Vegas, NV, USA, 4–7 March 2018; pp. 337–340. [Google Scholar]
  56. Pham, H.N.; Koay, C.Y.; Chakraborty, T.; Gupta, S.; Tan, B.L.; Wu, H.; Vardhan, A.; Nguyen, Q.H.; Palaparthi, N.R.; Nguyen, B.P.; et al. Lesion Segmentation and Automated Melanoma Detection using Deep Convolutional Neural Networks and XGBoost. In Proceedings of the 2019 International Conference on System Science and Engineering (ICSSE), Dong Hoi, Vietnam, 20–21 July 2019; pp. 142–147. [Google Scholar]
  57. Khouloud, S.; Ahlem, M.; Fadel, T.; Amel, S. W-net and inception residual network for skin lesion segmentation and classification. Appl. Intell. 2021, 52, 3976–3994. [Google Scholar] [CrossRef]
  58. Xie, Y.; Zhang, J.; Xia, Y.; Shen, C. A Mutual Bootstrapping Model for Automated Skin Lesion Segmentation and Classification. IEEE Trans. Med. Imaging 2020, 39, 2482–2493. [Google Scholar] [CrossRef]
  59. Yu, L.; Chen, H.; Dou, Q.; Qin, J.; Heng, P.A. Automated Melanoma Recognition in Dermoscopy Images via Very Deep Residual Networks. IEEE Trans. Med. Imaging 2017, 36, 994–1004. [Google Scholar] [CrossRef]
  60. Gouda, W.; Sama, N.U.; Al-Waakid, G.; Humayun, M.; Jhanjhi, N.Z. Detection of Skin Cancer Based on Skin Lesion Images Using Deep Learning. Healthcare 2022, 10, 1183. [Google Scholar] [CrossRef]
  61. Afza, F.; Sharif, M.; Mittal, M.; Khan, M.A.; Jude Hemanth, D. A hierarchical three-step superpixels and deep learning framework for skin lesion classification. Methods 2022, 88–102. [Google Scholar] [CrossRef]
  62. Benyahia, S.; Meftah, B.; Lézoray, O. Multi-features extraction based on deep learning for skin lesion classification. Tissue Cell 2022, 74, 101701. [Google Scholar] [CrossRef]
  63. Ameri, A. A Deep Learning Approach to Skin Cancer Detection in Dermoscopy Images. J. Biomed. Phys. Eng. 2020, 10, 801–806. [Google Scholar] [CrossRef] [PubMed]
  64. Zaman, K.; Bangash, J.I.; Maghdid, S.S.; Hassan, S.; Afridi, H.; Zohaib, M. Analysis and Classification of Skin Cancer Images using Convolutional Neural Network Approach. In Proceedings of the 2020 4th International Symposium on Multidisciplinary Studies and Innovative Technologies (ISMSIT), Istanbul, Turkey, 22–24 October 2020; pp. 1–8. [Google Scholar]
  65. Xiang, A.; Wang, F. Towards Interpretable Skin Lesion Classification with Deep Learning Models. AMIA Annu. Symp. Proc. 2019, 2019, 1246–1255. [Google Scholar] [PubMed]
  66. Tschandl, P.; Argenziano, G.; Razmara, M.; Yap, J. Diagnostic accuracy of content-based dermatoscopic image retrieval with deep classification features. Br. J. Dermatol. 2019, 181, 155–165. [Google Scholar] [CrossRef] [PubMed]
  67. Nauta, M.; Walsh, R.; Dubowski, A.; Seifert, C. Uncovering and Correcting Shortcut Learning in Machine Learning Models for Skin Cancer Diagnosis. Diagnostics 2021, 12, 40. [Google Scholar] [CrossRef] [PubMed]
  68. Rehman, M.Z.U.; Ahmed, F.; Alsuhibany, S.A.; Jamal, S.S.; Ali, M.Z.; Ahmad, J. Classification of Skin Cancer Lesions Using Explainable Deep Learning. Sensors 2022, 22, 6915. [Google Scholar] [CrossRef]
  69. Singh, R.K.; Gorantla, R.; Allada, S.G.R.; Narra, P. SkiNet: A deep learning framework for skin lesion diagnosis with uncertainty estimation and explainability. PLoS ONE 2022, 17, e0276836. [Google Scholar] [CrossRef]
  70. Alfi, I.A.; Rahman, M.; Shorfuzzaman, M.; Nazir, A. A Non-Invasive Interpretable Diagnosis of Melanoma Skin Cancer Using Deep Learning and Ensemble Stacking of Machine Learning Models. Diagnostics 2022, 12, 726. [Google Scholar] [CrossRef]
  71. Lee, J.R.H.; Pavlova, M.; Famouri, M.; Wong, A. Cancer-Net SCa: Tailored deep neural network designs for detection of skin cancer from dermoscopy images. BMC Med. Imaging 2022, 22, 143. [Google Scholar] [CrossRef]
  72. Wang, S.; Yin, Y.; Wang, D.; Wang, Y.; Jin, Y. Interpretability-Based Multimodal Convolutional Neural Networks for Skin Lesion Diagnosis. IEEE Trans. Cybern. 2022, 52, 12623–12637. [Google Scholar] [CrossRef]
  73. Maiti, R.; Agarwal, P.; Kumar, R.R.; Bhat, A. Detection of skin cancer using neural architecture search with model quantization. In Proceedings of the 2021 5th International Conference on Intelligent Computing and Control Systems (ICICCS), Madurai, India, 6–8 May 2021; pp. 1807–1814. [Google Scholar]
  74. Shinde, R.K.; Alam, S.; Hossain, B.; Imtiaz, S.M.; Kim, J.; Padwal, A.A.; Kim, N. Squeeze-MNet: Precise Skin Cancer Detection Model for Low Computing IoT Devices Using Transfer Learning. Cancers 2022, 15, 12. [Google Scholar] [CrossRef]
  75. Aishwarya, U.; Daniel, I.J.; Raghul, R. Convolutional Neural Network based Skin Lesion Classification and Identification. In Proceedings of the 2020 International Conference on Inventive Computation Technologies (ICICT), Coimbatore, India, 26–28 February 2020; pp. 264–270. [Google Scholar]
  76. Demir, A.; Yilmaz, F.; Kose, O. Early detection of skin cancer using deep learning architectures: Resnet-101 and inception-v3. In Proceedings of the 2019 Medical Technologies Congress (TIPTEKNO), Izmir, Turkey, 3–5 October 2019. [Google Scholar]
  77. Durupt, M.; Bouchy, O.; Christophe, S.; Kivits, J.; Boivin, J.-M. La télémédecine en zones rurales: Représentations et expériences de médecins généralistes. Sante Publique 2016, 28, 487–497. [Google Scholar] [CrossRef] [PubMed]
  78. Sanketh, R.S.; Bala, M.M.; Reddy, P.V.N.; Kumar, G.V.S.P. Melanoma Disease Detection Using Convolutional Neural Networks. In Proceedings of the 2020 4th International Conference on Intelligent Computing and Control Systems (ICICCS), Madurai, India, 13–15 May 2020; pp. 1031–1037. [Google Scholar]
  79. Shah, V.; Autee, P.; Sonawane, P. Detection of Melanoma from Skin Lesion Images using Deep Learning Techniques. In Proceedings of the 2020 International Conference on Data Science and Engineering (ICDSE), Kochi, India, 3–5 December 2020; pp. 1–8. [Google Scholar]
  80. Sherif, F.; Mohamed, W.A.; Mohra, A. Skin lesion analysis toward melanoma detection using deep learning techniques. Int. J. Electron. Telecommun. 2019, 65, 597–602. [Google Scholar] [CrossRef]
  81. Yang, S.; Shu, C.; Hu, H.; Ma, G.; Yang, M. Dermoscopic Image Classification of Pigmented Nevus under Deep Learning and the Correlation with Pathological Features. Comput. Math. Methods Med. 2022, 2022, 9726181. [Google Scholar] [CrossRef] [PubMed]
  82. Alwakid, G.; Gouda, W.; Humayun, M.; Sama, N.U. Melanoma Detection Using Deep Learning-Based Classifications. Healthcare 2022, 10, 2481. [Google Scholar] [CrossRef] [PubMed]
  83. Mahum, R.; Aladhadh, S. Skin Lesion Detection Using Hand-Crafted and DL-Based Features Fusion and LSTM. Diagnostics 2022, 12, 2974. [Google Scholar] [CrossRef]
  84. Alenezi, F.; Armghan, A.; Polat, K. A Novel Multi-Task Learning Network Based on Melanoma Segmentation and Classification with Skin Lesion Images. Diagnostics 2023, 13, 262. [Google Scholar] [CrossRef]
  85. Hsu, B.W.-Y.; Tseng, V.S. Hierarchy-aware contrastive learning with late fusion for skin lesion classification. Comput. Methods Programs Biomed. 2022, 216, 106666. [Google Scholar] [CrossRef]
  86. Akram, A.; Rashid, J.; Jaffar, M.A.; Faheem, M.; Amin, R.U. Segmentation and classification of skin lesions using hybrid deep learning method in the Internet of Medical Things. Ski. Res. Technol. 2023, 29, e13524. [Google Scholar] [CrossRef]
  87. Dascalu, A.; Walker, B.N.; Oron, Y.; David, E.O. Non-melanoma skin cancer diagnosis: A comparison between dermoscopic and smartphone images by unified visual and sonification deep learning algorithms. J. Cancer Res. Clin. Oncol. 2021, 148, 2497–2505. [Google Scholar] [CrossRef]
  88. Alzahrani, S.; Al-Nuaimy, W.; Al-Bander, B. Seven-Point Checklist with Convolutional Neural Networks for Melanoma Diagnosis. In Proceedings of the 2019 8th European Workshop on Visual Information Processing (EUVIP), Rome, Italy, 28–31 October 2019; pp. 211–216. [Google Scholar]
  89. Almufareh, M.F.; Tariq, N.; Humayun, M.; Khan, F.A. Melanoma identification and classification model based on fine-tuned convolutional neural network. Digit. Health 2024, 10, 20552076241253757. [Google Scholar] [CrossRef]
  90. Moturi, D.; Surapaneni, R.K.; Avanigadda, V.S.G. Developing an efficient method for melanoma detection using CNN techniques. J. Egypt. Natl. Cancer Inst. 2024, 36, 6. [Google Scholar] [CrossRef] [PubMed]
  91. Hossain, M.; Hossain, M.; Arefin, M.B.; Akhtar, F.; Blake, J. Combining State-of-the-Art Pre-Trained Deep Learning Models: A Noble Approach for Skin Cancer Detection Using Max Voting Ensemble. Diagnostics 2023, 14, 89. [Google Scholar] [CrossRef] [PubMed]
  92. Kwiatkowska, D.; Kluska, P.; Reich, A. Convolutional neural networks for the detection of malignant melanoma in dermoscopy images. Adv. Dermatol. Allergol. 2021, 38, 412–420. [Google Scholar] [CrossRef] [PubMed]
  93. Budhiman, A.; Suyanto, S.; Arifianto, A. Melanoma Cancer Classification Using ResNet with Data Augmentation. In Proceedings of the 2019 International Seminar on Research of Information Technology and Intelligent Systems (ISRITI), Yogyakarta, Indonesia, 5–6 December 2019; pp. 17–20. [Google Scholar]
  94. Hossin, A.; Rupom, F.F.; Mahi, H.R.; Sarker, A.; Ahsan, F.; Warech, S. Melanoma Skin Cancer Detection Using Deep Learning and Advanced Regularizer. In Proceedings of the 2020 International Conference on Advanced Computer Science and Information Systems (ICACSIS), Depok, Indonesia, 17–18 October 2020; pp. 89–94. [Google Scholar]
  95. Maiti, A.; Chatterjee, B. Improving detection of Melanoma and Naevus with deep neural networks. Multimedia Tools Appl. 2019, 79, 15635–15654. [Google Scholar] [CrossRef]
  96. Jayalakshmi, G.; Kumar, V.S. Performance analysis of Convolutional Neural Network (CNN) based Cancerous Skin Lesion Detection System. In Proceedings of the 2019 International Conference on Computational Intelligence in Data Science (ICCIDS), Gurugram, India, 6–7 September 2019. [Google Scholar]
  97. Seeja, R.D.; Suresh, A. Melanoma Segmentation and Classification using Deep Learning. Int. J. Innov. Technol. Explor. Eng. 2019, 8, 2667–2672. [Google Scholar] [CrossRef]
  98. Rokhana, R.; Herulambang, W.; Indraswari, R. Deep Convolutional Neural Network for Melanoma Image Classification. In Proceedings of the 2020 International Electronics Symposium (IES), Surabaya, Indonesia, 29–30 September 2020; pp. 481–486. [Google Scholar]
  99. Albahar, M.A. Skin Lesion Classification Using Convolutional Neural Network with Novel Regularizer. IEEE Access 2019, 7, 38306–38313. [Google Scholar] [CrossRef]
  100. Afza, F.; Sharif, M.; Khan, M.A.; Tariq, U.; Yong, H.-S.; Cha, J. Multiclass Skin Lesion Classification Using Hybrid Deep Features Selection and Extreme Learning Machine. Sensors 2022, 22, 799. [Google Scholar] [CrossRef]
  101. Guo, L.; Xie, G.; Xu, X.; Ren, J. Effective Melanoma Recognition Using Deep Convolutional Neural Network with Covariance Discriminant Loss. Sensors 2020, 20, 5786. [Google Scholar] [CrossRef]
  102. Mishra, V.; Kumar, A.; Arora, M. Deep convolution neural network based automatic multi-class classification of skin cancer from dermoscopic images. In Proceedings of the 2021 5th International Conference on Intelligent Computing and Control Systems (ICICCS), Madurai, India, 6–8 May 2021; pp. 800–805. [Google Scholar]
  103. Hagerty, J.R.; Stanley, R.J.; Almubarak, H.A.; Lama, N.; Kasmi, R.; Guo, P.; Drugge, R.J.; Rabinovitz, H.S.; Oliviero, M.; Stoecker, W.V. Deep Learning and Handcrafted Method Fusion: Higher Diagnostic Accuracy for Melanoma Dermoscopy Images. IEEE J. Biomed. Health Inform. 2019, 23, 1385–1391. [Google Scholar] [CrossRef]
  104. Ashraf, R.; Kiran, I.; Mahmood, T.; Butt, A.U.R.; Razzaq, N.; Farooq, Z. An efficient technique for skin cancer classification using deep learning. In Proceedings of the 2020 IEEE 23rd International Multitopic Conference (INMIC), Bahawalpur, Pakistan, 5–7 November 2020; pp. 1–5. [Google Scholar]
  105. Gouda, N.; Amudha, J. Skin Cancer Classification using ResNet. In Proceedings of the 2020 IEEE 5th International Conference on Computing Communication and Automation (ICCCA), Greater Noida, India, 30–31 October 2020; pp. 536–541. [Google Scholar]
  106. Dahou, A.; Aseeri, A.O.; Mabrouk, A.; Ibrahim, R.A.; Al-Betar, M.A.; Elaziz, M.A. Optimal Skin Cancer Detection Model Using Transfer Learning and Dynamic-Opposite Hunger Games Search. Diagnostics 2023, 13, 1579. [Google Scholar] [CrossRef]
  107. Li, G.; Jimenez, G. Optimal diagnosis of the skin cancer using a hybrid deep neural network and grasshopper optimization algorithm. Open Med. 2022, 17, 508–517. [Google Scholar] [CrossRef] [PubMed]
  108. Tahir, M.; Naeem, A.; Malik, H.; Tanveer, J.; Naqvi, R.A.; Lee, S.-W. DSCC_Net: Multi-Classification Deep Learning Models for Diagnosing of Skin Cancer Using Dermoscopic Images. Cancers 2023, 15, 2179. [Google Scholar] [CrossRef] [PubMed]
  109. Gilani, S.Q.; Syed, T.; Umair, M.; Marques, O. Skin Cancer Classification Using Deep Spiking Neural Network. J. Digit. Imaging 2023, 36, 1137–1147. [Google Scholar] [CrossRef] [PubMed]
  110. Wang, Y.; Su, J.; Xu, Q.; Zhong, Y. A Collaborative Learning Model for Skin Lesion Segmentation and Classification. Diagnostics 2023, 13, 912. [Google Scholar] [CrossRef]
  111. Raghavendra, P.V.S.P.; Charitha, C.; Begum, K.G.; Prasath, V.B.S. Deep Learning–Based Skin Lesion Multi-class Classification with Global Average Pooling Improvement. J. Digit. Imaging 2023, 36, 2227–2248. [Google Scholar] [CrossRef]
  112. Juan, C.-K.; Su, Y.-H.; Wu, C.-Y.; Yang, C.-S.; Hsu, C.-H.; Hung, C.-L.; Chen, Y.-J. Deep convolutional neural network with fusion strategy for skin cancer recognition: Model development and validation. Sci. Rep. 2023, 13, 17087. [Google Scholar] [CrossRef]
  113. Josphineleela, R.; Rao, P.B.V.R.; Shaikh, A.; Sudhakar, K. A Multi-Stage Faster RCNN-Based iSPLInception for Skin Disease Classification Using Novel Optimization. J. Digit. Imaging 2023, 36, 2210–2226. [Google Scholar] [CrossRef]
  114. Mehmood, A.; Gulzar, Y.; Ilyas, Q.M.; Jabbari, A.; Ahmad, M.; Iqbal, S. SBXception: A Shallower and Broader Xception Architecture for Efficient Classification of Skin Lesions. Cancers 2023, 15, 3604. [Google Scholar] [CrossRef]
  115. Radhika, V.; Chandana, B.S. MSCDNet-based multi-class classification of skin cancer using dermoscopy images. PeerJ Comput. Sci. 2023, 9, e1520. [Google Scholar] [CrossRef]
  116. Islam, K.; Ali, S.; Ali, M.; Haque, M.F.; Das, A.A.; Hossain, M.; Duranta, D.S.; Rahman, A. Melanoma Skin Lesions Classification using Deep Convolutional Neural Network with Transfer Learning. In Proceedings of the 2021 1st International Conference on Artificial Intelligence and Data Analytics (CAIDA), Riyadh, Saudi Arabia, 6–7 April 2021; pp. 48–53. [Google Scholar]
  117. Kanchana, K.; Kavitha, S.; Anoop, K.J.; Chinthamani, B. Enhancing Skin Cancer Classification using Efficient Net B0-B7 through Convolutional Neural Networks and Transfer Learning with Patient-Specific Data. Asian Pac. J. Cancer Prev. 2024, 25, 1795–1802. [Google Scholar] [CrossRef]
  118. Selvaraj, K.M.; Gnanagurusubbiah, S.; Roy, R.R.R.; Peter, J.H.J.; Balu, S. Enhancing skin lesion classification with advanced deep learning ensemble models: A path towards accurate medical diagnostics. Curr. Probl. Cancer 2024, 49, 101077. [Google Scholar] [CrossRef] [PubMed]
  119. Tang, X.; Sheykhahmad, F.R. Boosted dipper throated optimization algorithm-based Xception neural network for skin cancer diagnosis: An optimal approach. Heliyon 2024, 10, e26415. [Google Scholar] [CrossRef]
  120. Ali, M.U.; Khalid, M.; Alshanbari, H.; Zafar, A.; Lee, S.W. Enhancing Skin Lesion Detection: A Multistage Multiclass Convolutional Neural Network-Based Framework. Bioengineering 2023, 10, 1430. [Google Scholar] [CrossRef] [PubMed]
  121. Hilmy, M.A.; Sasongko, P.S. Ensembles of Convolutional Neural Networks for Skin Lesion Dermoscopy Images Classification. In Proceedings of the 2019 3rd International Conference on Informatics and Computational Sciences (ICICoS), Semarang, Indonesia, 29–30 October 2019; pp. 1–6. [Google Scholar]
  122. Jusman, Y.; Firdiantika, I.M.; Dharmawan, D.A.; Purwanto, K. Performance of Multi Layer Perceptron and Deep Neural Networks in Skin Cancer Classification. In Proceedings of the 2021 IEEE 3rd Global Conference on Life Sciences and Technologies (LifeTech), Kyoto, Japan, 10–12 March 2021; pp. 534–538. [Google Scholar]
  123. Jeny, A.A.; Sakib, A.N.M.; Junayed, M.S.; Lima, K.A.; Ahmed, I.; Islam, B. SkNet: A Convolutional Neural Networks Based Classification Approach for Skin Cancer Classes. In Proceedings of the 2020 23rd International Conference on Computer and Information Technology (ICCIT), Dhaka, Bangladesh, 19–21 December 2020; pp. 1–6. [Google Scholar]
  124. Khan, M.A.; Zhang, Y.-D.; Sharif, M.; Akram, T. Pixels to Classes: Intelligent Learning Framework for Multiclass Skin Lesion Localization and Classification. Comput. Electr. Eng. 2021, 90, 106956. [Google Scholar] [CrossRef]
  125. Li, Y.; Shen, L. Skin Lesion Analysis towards Melanoma Detection Using Deep Learning Network. Sensors 2018, 18, 556. [Google Scholar] [CrossRef] [PubMed]
  126. Hosny, K.M.; Kassem, M.A. Refined Residual Deep Convolutional Network for Skin Lesion Classification. J. Digit. Imaging 2022, 35, 258–280. [Google Scholar] [CrossRef]
  127. Reis, H.C.; Turk, V.; Khoshelham, K.; Kaya, S. InSiNet: A deep convolutional approach to skin cancer detection and segmentation. Med. Biol. Eng. Comput. 2022, 60, 643–662. [Google Scholar] [CrossRef]
  128. Reisinho, J.; Coimbra, M.; Renna, F. Deep Convolutional Neural Network Ensembles For Multi-Classification of Skin Lesions From Dermoscopic and Clinical Images. In Proceedings of the 2020 42nd Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) in Conjunction with the 43rd Annual Conference of the Canadian Medical and Biological Engineering Society, Montreal, QC, Canada, 20–24 July 2020; pp. 1940–1943. [Google Scholar]
  129. Al-Masni, M.A.; Kim, D.-H.; Kim, T.-S. Multiple skin lesions diagnostics via integrated deep convolutional networks for segmentation and classification. Comput. Methods Programs Biomed. 2020, 190, 105351. [Google Scholar] [CrossRef]
  130. Jasil, S.P.G.; Ulagamuthalvi, V. Deep learning architecture using transfer learning for classification of skin lesions. J. Ambient. Intell. Humaniz. Comput. 2021, 1–8. [Google Scholar] [CrossRef]
  131. Jasil, S.G.; Ulagamuthalvi, V. Skin Lesion Classification Using Pre-Trained DenseNet201 Deep Neural Network. In Proceedings of the 2021 3rd International Conference on Signal Processing and Communication (ICPSC), Coimbatore, India, 13–14 May 2021; pp. 393–396. [Google Scholar]
  132. Sevli, O. A deep convolutional neural network-based pigmented skin lesion classification application and experts evaluation. Neural Comput. Appl. 2021, 33, 12039–12050. [Google Scholar] [CrossRef]
  133. Rezvantalab, A.; Safigholi, H.; Karimijeshni, S. Dermatologist Level Dermoscopy Skin Cancer Classification Using Different Deep Learning Convolutional Neural Networks Algorithms. 2018. Available online: http://arxiv.org/abs/1810.10348 (accessed on 8 October 2024).
  134. Chaturvedi, S.S.; Tembhurne, J.V.; Diwan, T. A multi-class skin Cancer classification using deep convolutional neural networks. Multimedia Tools Appl. 2020, 79, 28477–28498. [Google Scholar] [CrossRef]
  135. Haenssle, H.A.; Fink, C.; Schneiderbauer, R.; Toberer, F.; Buhl, T.; Blum, A.; Kalloo, A.; Hassen, A.B.H.; Thomas, L.; Enk, A.; et al. Man against machine: Diagnostic performance of a deep learning convolutional neural network for dermoscopic melanoma recognition in comparison to 58 dermatologists. Ann. Oncol. Off. J. Eur. Soc. Med. Oncol. 2018, 29, 1836–1842. [Google Scholar] [CrossRef] [PubMed]
  136. Fink, C.; Blum, A.; Buhl, T.; Mitteldorf, C.; Hofmann-Wellenhof, R.; Deinlein, T.; Stolz, W.; Trennheuser, L.; Cussigh, C.; Deltgen, D.; et al. Diagnostic performance of a deep learning convolutional neural network in the differentiation of combined naevi and melanomas. J. Eur. Acad. Dermatol. Venereol. 2019, 34, 1355–1361. [Google Scholar] [CrossRef] [PubMed]
  137. Fujisawa, Y.; Otomo, Y.; Ogata, Y.; Nakamura, Y.; Fujita, R.; Ishitsuka, Y.; Watanabe, R.; Okiyama, N.; Ohara, K.; Fujimoto, M. Deep-learning-based, computer-aided classifier developed with a small dataset of clinical images surpasses board-certified dermatologists in skin tumour diagnosis. Br. J. Dermatol. 2018, 180, 373–381. [Google Scholar] [CrossRef]
  138. Jinnai, S.; Yamazaki, N.; Hirano, Y.; Sugawara, Y.; Ohe, Y.; Hamamoto, R. The Development of a Skin Cancer Classification System for Pigmented Skin Lesions Using Deep Learning. Biomolecules 2020, 10, 1123. [Google Scholar] [CrossRef]
  139. Hekler, A.; Utikal, J.S.; Enk, A.H.; Hauschild, A.; Weichenthal, M.; Maron, R.C.; Berking, C.; Haferkamp, S.; Klode, J.; Schadendorf, D.; et al. Superior skin cancer classification by the combination of human and artificial intelligence. Eur. J. Cancer 2019, 120, 114–121. [Google Scholar] [CrossRef]
  140. Aburaed, N.; Panthakkan, A.; Al-Saad, M.; Amin, S.A.; Mansoor, W. Deep Convolutional Neural Network (DCNN) for Skin Cancer Classification. In Proceedings of the 2020 27th IEEE International Conference on Electronics, Circuits and Systems (ICECS), Glasgow, UK, 23–25 November 2020; pp. 3–6. [Google Scholar]
  141. ElGhany, S.A.; Ibraheem, M.R.; Alruwaili, M.; Elmogy, M. Diagnosis of Various Skin Cancer Lesions Based on Fine-Tuned ResNet50 Deep Network. Comput. Mater. Contin. 2021, 68, 117–135. [Google Scholar] [CrossRef]
  142. Emara, T.; Afify, H.M.; Ismail, F.H.; Hassanien, A.E. A Modified Inception-v4 for Imbalanced Skin Cancer Classification Dataset. In Proceedings of the 2019 14th International Conference on Computer Engineering and Systems (ICCES), Cairo, Egypt, 17–18 December 2019; pp. 28–33. [Google Scholar]
  143. Gessert, N.; Sentker, T.; Madesta, F.; Schmitz, R.; Kniep, H.; Baltruschat, I.; Werner, R.; Schlaefer, A. Skin Lesion Classification Using CNNs with Patch-Based Attention and Diagnosis-Guided Loss Weighting. IEEE Trans. Biomed. Eng. 2019, 67, 495–503. [Google Scholar] [CrossRef]
  144. Kondaveeti, H.K.; Edupuganti, P. Skin Cancer Classification using Transfer Learning. In Proceedings of the 2020 IEEE International Conference on Advent Trends in Multidisciplinary Research and Innovation (ICATMRI), Buldhana, India, 30 December 2020; pp. 1–4. [Google Scholar]
  145. Namozov, A.; Ergashev, D.; Cho, Y.I. Adaptive Activation Functions for Skin Lesion Classification Using Deep Neural Networks. In Proceedings of the 2018 Joint 10th International Conference on Soft Computing and Intelligent Systems (SCIS) and 19th International Symposium on Advanced Intelligent Systems (ISIS), Toyama, Japan, 5–8 December 2018; pp. 232–235. [Google Scholar]
  146. Pai, K.; Giridharan, A. Convolutional Neural Networks for classifying skin lesions. In Proceedings of the TENCON 2019—2019 IEEE Region 10 Conference (TENCON), Kochi, India, 17–20 October 2019; pp. 1794–1796. [Google Scholar]
  147. Salian, A.C.; Vaze, S.; Singh, P.; Shaikh, G.N.; Chapaneri, S.; Jayaswal, D. Skin Lesion Classification using Deep Learning Architectures. In Proceedings of the 2020 3rd International Conference on Communication System, Computing and IT Applications (CSCITA), Mumbai, India, 3–4 April 2020; Volume 703, pp. 168–173. [Google Scholar]
  148. Subramanian, R.R.; Achuth, D.; Kumar, P.S.; Reddy, K.N.K.; Amara, S.; Chowdary, A.S. Skin cancer classification using Convolutional neural networks. In Proceedings of the 2021 11th International Conference on Cloud Computing, Data Science & Engineering (Confluence), Noida, India, 28–29 January 2021; pp. 13–19. [Google Scholar]
  149. Thurnhofer-Hemsi, K.; Lopez-Rubio, E.; Dominguez, E.; Elizondo, D.A. Skin Lesion Classification by Ensembles of Deep Convolutional Networks and Regularly Spaced Shifting. IEEE Access 2021, 9, 112193–112205. [Google Scholar] [CrossRef]
  150. Fraiwan, M.; Faouri, E. On the Automatic Detection and Classification of Skin Cancer Using Deep Transfer Learning. Sensors 2022, 22, 4963. [Google Scholar] [CrossRef]
  151. Naeem, A.; Anees, T.; Fiza, M.; Naqvi, R.A.; Lee, S.-W. SCDNet: A Deep Learning-Based Framework for the Multiclassification of Skin Cancer Using Dermoscopy Images. Sensors 2022, 22, 5652. [Google Scholar] [CrossRef] [PubMed]
  152. Arshad, M.; Khan, M.A.; Tariq, U.; Armghan, A.; Alenezi, F.; Javed, M.Y.; Aslam, S.M.; Kadry, S. A Computer-Aided Diagnosis System Using Deep Learning for Multiclass Skin Lesion Classification. Comput. Intell. Neurosci. 2021, 2021, 9619079. [Google Scholar] [CrossRef] [PubMed]
  153. Alkarakatly, T.; Eidhah, S.; Al-Sarawani, M.; Al-Sobhi, A.; Bilal, M. Skin Lesions Identification Using Deep Convolutional Neural Network. In Proceedings of the 2019 International Conference on Advances in the Emerging Computing Technologies (AECT), Medina, Saudi Arabia, 13–15 May 2020; pp. 1–5. [Google Scholar]
  154. Hosny, K.M.; Kassem, M.A.; Foaud, M.M. Skin Cancer Classification using Deep Learning and Transfer Learning. In Proceedings of the 2018 9th Cairo International Biomedical Engineering Conference (CIBEC), Cairo, Egypt, 20–22 December 2018; pp. 90–93. [Google Scholar]
  155. Hosny, K.M.; Kassem, M.A.; Foaud, M.M. Classification of skin lesions using transfer learning and augmentation with Alex-net. PLoS ONE 2019, 14, e0217293. [Google Scholar] [CrossRef] [PubMed]
  156. Serener, A.; Serte, S. Keratinocyte Carcinoma Detection via Convolutional Neural Networks. In Proceedings of the 2019 3rd International Symposium on Multidisciplinary Studies and Innovative Technologies (ISMSIT), Ankara, Turkey, 11–13 October 2019; pp. 1–5. [Google Scholar]
  157. Serte, S.; Demirel, H. Gabor wavelet-based deep learning for skin lesion classification. Comput. Biol. Med. 2019, 113, 103423. [Google Scholar] [CrossRef] [PubMed]
  158. Iqbal, I.; Younus, M.; Walayat, K.; Kakar, M.U.; Ma, J. Automated multi-class classification of skin lesions through deep convolutional neural network with dermoscopic images. Comput. Med. Imaging Graph. 2021, 88, 101843. [Google Scholar] [CrossRef]
  159. Alkarakatly, T.; Eidhah, S.; Al-Sarawani, M.; Al-Sobhi, A.; Bilal, M. Skin lesions identification using deep convolutional neural network. IEEE Access 2020, 8, 264–270. [Google Scholar] [CrossRef]
  160. Kassem, M.A.; Hosny, K.M.; Fouad, M.M. Skin Lesions Classification Into Eight Classes for ISIC 2019 Using Deep Convolutional Neural Network and Transfer Learning. IEEE Access 2020, 8, 114822–114832. [Google Scholar] [CrossRef]
  161. Ahmed, S.A.A.; Yanikoglu, B.; Goksu, O.; Aptoula, E. Skin Lesion Classification with Deep CNN Ensembles. In Proceedings of the 2020 28th Signal Processing and Communications Applications Conference (SIU), Gaziantep, Turkey, 5–7 October 2020; pp. 1–4. [Google Scholar]
  162. Barata, C.; Marques, J.S. Deep Learning For Skin Cancer Diagnosis with Hierarchical Architectures. In Proceedings of the 2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI), Venice, Italy, 8–11 April 2019; pp. 841–845. [Google Scholar]
  163. Kaymak, S.; Esmaili, P.; Serener, A. Deep Learning for Two-Step Classification of Malignant Pigmented Skin Lesions. In Proceedings of the 14th Symposium on Neural Networks and Applications (NEUREL 2018), Belgrade, Serbia, 20–21 November 2018; pp. 1–6. [Google Scholar] [CrossRef]
  164. Moldovan, D. Transfer Learning Based Method for Two-Step Skin Cancer Images Classification. In Proceedings of the 2019 E-Health and Bioengineering Conference (EHB), Iasi, Romania, 21–23 November 2019; pp. 1–4. [Google Scholar]
  165. Han, S.S.; Kim, M.S.; Lim, W.; Park, G.H.; Park, I.; Chang, S.E. Classification of the Clinical Images for Benign and Malignant Cutaneous Tumors Using a Deep Learning Algorithm. J. Investig. Dermatol. 2018, 138, 1529–1538. [Google Scholar] [CrossRef]
  166. Pratiwi, R.A.; Nurmaini, S.; Rini, D.P.; Rachmatullah, M.N.; Darmawahyuni, A. Deep ensemble learning for skin lesions classification with convolutional neural network. IAES Int. J. Artif. Intell. (IJ-AI) 2021, 10, 563. [Google Scholar] [CrossRef]
  167. Guo, Y.; Ashour, A.S.; Si, L.; Mandalaywala, D.P. Multiple Convolutional Neural Network for Skin Dermoscopic Image Classification. In Proceedings of the 2018 IEEE International Symposium on Signal Processing and Information Technology (ISSPIT), Louisville, KY, USA, 6–8 December 2018; pp. 365–369. [Google Scholar]
  168. Shahin, A.H.; Kamal, A.; Elattar, M.A. Deep Ensemble Learning for Skin Lesion Classification from Dermoscopic Images. In Proceedings of the 2018 9th Cairo International Biomedical Engineering Conference (CIBEC), Cairo, Egypt, 20–22 December 2018; pp. 150–153. [Google Scholar]
  169. Nugroho, E.S.; Ardiyanto, I.; Nugroho, H.A. Boosting the performance of pretrained CNN architecture on dermoscopic pigmented skin lesion classification. Ski. Res. Technol. 2023, 29, e13505. [Google Scholar] [CrossRef]
  170. Obayya, M.; Arasi, M.A.; Almalki, N.S.; Alotaibi, S.S.; Al Sadig, M.; Sayed, A. Internet of Things-Assisted Smart Skin Cancer Detection Using Metaheuristics with Deep Learning Model. Cancers 2023, 15, 5016. [Google Scholar] [CrossRef] [PubMed]
  171. Hussain, M.; Khan, M.A.; Damaševičius, R.; Alasiry, A.; Marzougui, M.; Alhaisoni, M.; Masood, A. SkinNet-INIO: Multiclass Skin Lesion Localization and Classification Using Fusion-Assisted Deep Neural Networks and Improved Nature-Inspired Optimization Algorithm. Diagnostics 2023, 13, 2869. [Google Scholar] [CrossRef] [PubMed]
  172. Lilhore, U.K.; Simaiya, S.; Sharma, Y.K.; Kaswan, K.S.; Rao, K.B.V.B.; Rao, V.V.R.M.; Baliyan, A.; Bijalwan, A.; Alroobaea, R. A precise model for skin cancer diagnosis using hybrid U-Net and improved MobileNet-V3 with hyperparameters optimization. Sci. Rep. 2024, 14, 4299. [Google Scholar] [CrossRef]
  173. Nguyen, V.D.; Bui, N.D.; Do, H.K. Skin Lesion Classification on Imbalanced Data Using Deep Learning with Soft Attention. Sensors 2022, 22, 7530. [Google Scholar] [CrossRef]
  174. Jaisakthi, S.M.; Mirunalini, P.; Aravindan, C.; Appavu, R. Classification of skin cancer from dermoscopic images using deep neural network architectures. Multimed. Tools Appl. 2022, 82, 15763–15778. [Google Scholar] [CrossRef]
  175. Mehr, R.A.; Ameri, A. Skin Cancer Detection Based on Deep Learning. J. Biomed. Phys. Eng. 2022, 12, 559–568. [Google Scholar] [CrossRef]
  176. Pacheco, A.G.C.; Krohling, R.A. An Attention-Based Mechanism to Combine Images and Metadata in Deep Learning Models Applied to Skin Cancer Classification. IEEE J. Biomed. Health Inform. 2021, 25, 3554–3563. [Google Scholar] [CrossRef]
  177. Ou, C.; Zhou, S.; Yang, R.; Jiang, W.; He, H.; Gan, W.; Chen, W.; Qin, X.; Luo, W.; Pi, X.; et al. A deep learning based multimodal fusion model for skin lesion diagnosis using smartphone collected clinical images and metadata. Front. Surg. 2022, 9, 1029991. [Google Scholar] [CrossRef]
  178. Mijwil, M.M. Skin cancer disease images classification using deep learning solutions. Multimedia Tools Appl. 2021, 80, 1–17. [Google Scholar] [CrossRef]
Figure 1. PRISMA flow diagram for the systematic review of deep learning applications for skin cancer diagnosis [16].
Figure 1. PRISMA flow diagram for the systematic review of deep learning applications for skin cancer diagnosis [16].
Biomedinformatics 04 00121 g001
Table 1. Records included in the systematic review for binary classification and respective network architecture, employed database, sample size, training/testing/validation split, ACC SN and SP values.
Table 1. Records included in the systematic review for binary classification and respective network architecture, employed database, sample size, training/testing/validation split, ACC SN and SP values.
Year [Ref.]Network 1DatabaseDataset SizeTraining/
Testing/
Validation Split
ACCSNSP
2016 [50]CNNMED-DONE170-0.810.810.80
2016 [2,59]CNNISIC 201612500.72/0.28/-0.8550.5070.94
2017 [3,36]VGG-16ISIC 201726000.77/0.23/-0.7970.3410.907
2018 [4,167]ResNet-50ISIC 201612790.7/0.3/-0.852--
2018 [5,18]Google-, VGG- Alex-NetISIC 201726000.77/0.23/-0.838--
2018 [6,135]Inception-V4ISIC 2016300--0.8890.825
2018 [7,154]AlexNetPH2200-0.9860.9830.989
2018 [8,163]AlexNetHAM1000010,0150.8/0.2/-0.780.8390.635
2018 [9,55]CNNISIC 2016---0.960.89
2019 [137]GoogleNetPrivate dataset60090.8/0.2/-0.7650.9630.895
2019 [99]CNNISIC archives23,9060.7/0.3/-0.9750.9430.936
2019 [88]CNN--0.6/0.4/-0.6430.5530.892
2019 [76]Inception_V3ISIC archives30970.78/0.22/-0.9--
2019 [155]AlexNetMED-NODE170-0.9770.9730.973
2019 [96]CNNISIC archives24000.7/0.3/-0.893--
2019 [97]VGG-16ISIC 20169000.9/0.1/-0.9310.9550.962
2019 [56]DenseNet-161ISIC 201727500.72/0.22/0.060.863--
2019 [80]CNNISIC 20187000.7/0.3/-0.966--
2019 [103]ResNet-50---0.94--
2019 [157]CNN
2019 [58]CNNISIC archives, PH2--0.8780.7270.915
2019 [29]CNNISIC 2019 -0.963--
2020 [46]AlexNetDermis, DermQuest2220.77/0.23/-0.974--
2020 [136]CNNPrivate database72--0.9710.788
2020 [101]ResNetISIC 201810,0150.5/0.25/0.250.7690.9420.747
2020 [75]CNNISIC archives--0.9730.9870.998
2020 [94]CNNISIC 2019--0.935--
2020 [138]CNNPrivate dataset-0.89/0.11/-0.915--
2020 [24]ResNet-50, GoogleNet, AlexNetMED-NODE, Dermis and DermQuest376-0.93--
2020 [51]ResNet-101HAM1000010,015-0.977--
2020 [49]AlexNetISIC archives--0.8080.6610.896
2020 [98]CNNISIC archives32970.89/0.11/-0.8470.9190.787
2020 [78]CNNISIC archives27220.7/0.3/-0.91--
2020 [79]ResNet-50ISIC 20203455-0.9390.9970.556
2020 [26]VGG-16ISIC 201612760.7/0.3/-0.917--
2020 [101]CNNISIC 201810,0150.5/0.25/0.250.7690.9420.747
2021 [87]CNNHAM10000 and private dataset363-0.8780.9550.576
2021 [116]CNNHAM1000010,0150.8/0.2/-0.909--
2021 [95]CNNISIC archives21700.8/0.2/-0.968--
2021 [22]Efficient-B5ISIC 2020-0.7/0.2/0.1---
2021 [47]ResNet-152ISIC 201727420.72/0.21/0.070.8720.820.925
2021 [178]Inception_V3ISIC archives24,2250.8/0.2/-0.8690.8610.876
2021 [25]DenseNet-201ISIC 2020-----
2022 [60]CNNISIC 201811,527-0.8320.7530.897
2022 [68]MobileNetV2Kaggle73270.7/0.2/0.10.9180.910.913
2022 [70]MobileNet, Xception, ResNet50/50V2, DenseNet12ISIC 201832970.8/0.2/-0.920.920.9
2022 [71]CNNISIC archives23,9060.72/0.19/0.090.8440.9280.746
2022 [126]CNNISIC-201727500.75/0.15/0.150.9620.9410.972
2022 [107]AlexNetPH22000.8/0.2/-0.980.930.98
2022 [83]CNNISIC-201610000.75/0.25/-0.9940.9860.992
2022 [67]VGG-16ISIC archives21,8040.870.2/----
2023 [74]CNNISIC archives6550.8/0.2/-0.9380.9250.955
2023 [31]CNNMultiple databases36,703----
2023 [109]VGG-13ISIC 201914000.7/0.15/0.150.8960.8940.897
2023 [84]CNNISIC archives35620.8/0.2/-0.9770.9560.998
2023 [86]ResNet50ISIC 202028050.85/0.15/-0.967--
2023 [28]Xception, MobileNetV2Private database4080.75/0.15/0.10.9751-
2023 [54]CNNISIC-20172000-0.9930.990.993
2024 [91]EnsembleHAM1000010,015-0.932--
2024 [89]DenseNetISIC-201925,000-0.970.904-
2024 [90]CNNHAM1000010,015-0.95--
1 When no known CNN model is specified by the authors, a broad, original CNN is considered. If a known network model is indicated, it is considered to have been used as a foundation for the classification network. Consultation of the original research paper is advised for additional information.
Table 2. Records included in the systematic review for binary classification and respective network architecture, employed database, sample size, training/testing/validation split, ACC SN and SP values.
Table 2. Records included in the systematic review for binary classification and respective network architecture, employed database, sample size, training/testing/validation split, ACC SN and SP values.
Year [Ref.]NetworkDatabaseDataset SizeTraining/
Testing/
Validation Split
ACCSNSP
2018 [17]Google-, VGG- Alex-and Res-NetISIC 201726000.77/0.23/-0.8660.5560.785
2018 [165]ResNet-152Asan, MED-NODE and atlas site19,398----
2018 [125]CNNISIC 20172000-0.912--
2018 [145]LeNetISIC 2018--0.9586--
2018 [133]DenseNet_201, ResNet-152, Inception_V3, InceptionResNet_V2HAM10000, PH210,135----
2019 [168]ResNet-50, Inception_V3ISIC 201810,0150.8/0.2/-0.899--
2019 [65]VGG-16, DenseNet, Xception and InceptionResNet_V2HAM1000089170.6/0.2/0.20.856--
2019 [134]ResNeXt-101HAM1000010,0150.88/0.12/-0.932--
2019 [162]DenseNet-161ISIC 201727500.72/0.22/0.060.7
2019 [142]Inception_V4HAM1000010,0150.9/0.1/-0.947--
2019 [139]ResNet-50HAM1000, ISIC archives12,336--0.890.84
2019 [121]CNNHAM1000010,015-0.877--
2019 [32]MobileNetHAM1000010,0150.85/0.10/0.050.927-0.97
2019 [33]Inception-ResNetHAM1000010,0150.85/0.13/0.020.839--
2019 [164]CNNHAM1000010,015-0.85--
2019 [146]CNNHAM1000010,0150.8/0.2/-0.795--
2019 [156]ResNet-50ISIC 201984730.83/0.17/-0.870.850.9
2019 [66]ResNet-50EDRA, ISIC 2017, Private database20,3290.8/0.2/-0.768--
2020 [128]ResNet, Inception_V3, InceptionResNet_V2ISIC archives, DermNetNZ, Hellenic Dermatological Atlas, Danderm4000-0.7990.7990.933
2020 [129]ResNet-50ISIC 2016, 2017, 201810,0150.72/0.2/0.080.8930.810.872
2020 [161]Xception, InceptionResNet_V2 and NasNetLargeISIC 201925,331-0.937--
2020 [140]VGG-16HAM1000071820.8/0.16/0.02 of training data0.99--
2020 [153]CNNPH22000.72/0.2/0.080.950.940.97
2020 [160]GoogleNetISIC 201925,3310.8/0.1/0.10.9490.7980.97
2020 [104]CNN-12660.68/0.19/0.130.932--
2020 [63]AlexNetHAM1000034000.9/0.1/-0.840.810.88
2020 [93]ResNet-50ISIC 201821,6590.8/0.1/0.10.785--
2020 [130]VGG-16ISIC 201830910.8/0.2/-0.77--
2020 [105]ResNet-34ISIC 201925,331-0.92--
2020 [143]Dense121HAM1000010,015--0.7570.96
2020 [23]VGG-19HAM1000010,015-0.99--
2020 [123]SkNetPrivate dataset and online images48000.8/0.2/-0.952--
2020 [144]ResNet-50HAM1000010,015-0.9--
2020 [42]ResNet-50ISIC 201810,015-0.9520.8320.743
2020 [158]CNNISIC 20172750--0.930.91
2020 [34]CNNISIC archives--0.794--
2020 [35]DenseNet-121HAM1000010,0150.9/0.1/-0.92--
2020 [45]CNNHAM1000010,015-0.8840.7670.963
2020 [64]InceptionOnline images31500.75/0.25/-0.988--
2020 [44]DenseNetISIC 2016, 2017, 201827500.72/0.28/-0.857--
2021 [132]CNNHAM1000010,0150.8/0.1/0.10.915--
2021 [92]ResNetXHAM1000010,0150.81/0.08/0.11-0.830.9
2021 [166]Inception_V3, ResNet_V2, DenseNet-201HAM1000010,015-0.9720.9010.977
2021 [141]ResNet-50HAM1000010,0150.7/0.3/-0.9--
2021 [176]ResNet-50ISIC 2019, PAD-UFES-2027,629-0.909--
2021 [131]DenseNet-201ISIC 201830910.8/0.2/-0.77--
2021 [122]VGG-16HAM1000010,015-0.975--
2021 [73]EfficientNet-B3HAM1000010,015----
2021 [102]VGG-16ISIC archives32160.8/0.2/-0.796--
2021 [148]CNNHAM1000010,015-0.83--
2021 [124]CNNHAM1000010,015-0.902--
2021 [149]MobileNetV2, GoogleNetHAM1000010,015-0.8350.6560.954
2021 [152]ResNetHAM1000010,015-0.95--
2021 [57]Inception-ResNetISIC 2016, 2017, 2018, PH263940.75/0.25/-0.9810.9810.981
2021 [37]CNNHAM1000010,015-0.904--
2022 [19]VGG-16ISIC archive32970.85/0.10/0.050.890--
2022 [20]AlexNetISIC archive24000.8/0.2/-0.8710.800.942
2022 [38]RegNetY-320HAM1000010,015-0.91--
2022 [39]SCNN_12ISIC archive16,485-0.9880.9800.989
2022 [61]ResNet-50PH2200-0.954--
2022 [30]CNNISIC-20195635-0.688--
2022 [62]DenseNet201ISIC 201925,331-0.9230.8520.963
2022 [52]GC-SCNNISIC 201811,5270.8/0.2/-0.99711
2022 [53]XceptionNetHAM1000010,0150.8/0.2/-10.940.97
2022 [69]BayesianDenseNet-169ISIC 201810,015-0.873--
2022 [72]CNNHAM1000010,015-0.9510.8350.932
2022 [100]DNNISIC 201812,500-0.943--
2022 [177]DNNPrivate database2298-0.768--
2022 [174]EfficientNetISIC-2019, 202058,457----
2022 [85]CNNISIC-201911,527-0.8710.8420.889
2022 [82]Resnet-50HAM1000010,015-0.860.86-
2022 [81]CNNPrivate database268-0.8580.8860.902
2022 [175]Inception-ResNet-v2ISIC archives66,735-0.893--
2022 [173]InceptionResNetV2HAM1000010,015-0.90.81-
2022 [150]DenseNet201HAM1000010,0150.9/0.1/-0.8290.7360.96
2022 [151]VGG-16ISIC-201925,3310.7/0.2/0.10.9690.921-
2022 [127]InSiNetISIC-201810,0150.9/0.05/0.050.9450.9750.912
2022 [43]CNNHAM1001510,0150.9/0.1/-0.97--
2022 [110]CNNISIC-201727500.72/0.21/0.070.9070.708
2023 [21]GoogleNet, DarkNetISIC-201925,331-0.824--
2023 [147]CNN
2023 [111]CNNISIC archives--0.972--
2023 [106]MobileNetV3PH2200-0.9640.974-
2023 [108]CNNISIC 2020, HAM10000DermIS8012-0.9410.937-
2023 [169]InceptionV3ISIC-201925,331-0.9630.9110.986
2023 [170]CNNISIC archives308-0.9950.9770.997
2023 [112]CNNPrivate database463-0.850.820.93
2023 [171]CNNISIC201810,015-0.9610.969-
2023 [113]CNNHAM1000010,0150.8/0.2/-0.9580.965-
2023 [27]MobileNetV3PH22000.85/0.15/-0.967--
2023 [114]XceptionHAM1000010,0150.8/0.2/-0.9690.954-
2023 [40]CNNPH2, HAM10000, ISBI-201724,000-0.9560.9670.95
2023 [48]CNNPrivate database10,0150.8/0.15/0.050.7990.704-
2023 [115]CNNISIC-201925,3310.7/0.2/0.10.9870.984-
2024 [41]CNNPAD-UFES-201314-0.96--
2024 [120]CNNISIC201810,015-0.975--
2024 [117]EfficientNet-B7ISIC2016379-0.971--
2024 [118]ResNet, DenseNet201, GoogLeNet, and XceptionHAM1000010,015-0.97--
2024 [119]XceptionISIC-201925,331-0.9420.971-
2024 [172]MobileNet-V3HAM1000010,015-0.9890.9640.973
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Magalhaes, C.; Mendes, J.; Vardasca, R. Systematic Review of Deep Learning Techniques in Skin Cancer Detection. BioMedInformatics 2024, 4, 2251-2270. https://doi.org/10.3390/biomedinformatics4040121

AMA Style

Magalhaes C, Mendes J, Vardasca R. Systematic Review of Deep Learning Techniques in Skin Cancer Detection. BioMedInformatics. 2024; 4(4):2251-2270. https://doi.org/10.3390/biomedinformatics4040121

Chicago/Turabian Style

Magalhaes, Carolina, Joaquim Mendes, and Ricardo Vardasca. 2024. "Systematic Review of Deep Learning Techniques in Skin Cancer Detection" BioMedInformatics 4, no. 4: 2251-2270. https://doi.org/10.3390/biomedinformatics4040121

APA Style

Magalhaes, C., Mendes, J., & Vardasca, R. (2024). Systematic Review of Deep Learning Techniques in Skin Cancer Detection. BioMedInformatics, 4(4), 2251-2270. https://doi.org/10.3390/biomedinformatics4040121

Article Metrics

Back to TopTop