Next Article in Journal
Designing Comprehensive Shifting Control Strategy of Hydro-Mechanical Continuously Variable Transmission
Next Article in Special Issue
Quantitative Ultrasound Analysis of Oral Mucosa: An Observational Cross-Sectional Study
Previous Article in Journal
Skin Cancer Disease Detection Using Transfer Learning Technique
Previous Article in Special Issue
Photodynamic Therapy Using 5-Aminolevulinic Acid (Ala) for the Treatment of Chronic Periodontitis: A Prospective Case Series
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Applied Sciences
Volume 12
Issue 11
10.3390/app12115715
applsci-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Review

Machine-Learning Applications in Oral Cancer: A Systematic Review

by
Xaviera A. López-Cortés
1,*,
Felipe Matamala
1,
Bernardo Venegas
2 and
César Rivera
3
1
Department of Computer Sciences and Industries, Universidad Católica del Maule, Talca 3480112, Chile
2
Department of Stomatology, Faculty of Health Sciences, Universidad de Talca, Talca 3460000, Chile
3
Department of Basic Biomedical Sciences, Faculty of Health Sciences, Universidad de Talca, Talca 3460000, Chile
*
Author to whom correspondence should be addressed.
Appl. Sci. 2022, 12(11), 5715; https://doi.org/10.3390/app12115715
Submission received: 7 December 2021 / Revised: 20 May 2022 / Accepted: 31 May 2022 / Published: 4 June 2022
(This article belongs to the Special Issue Oral Pathology and Medicine: Diagnosis and Therapy)
Browse Figures
Versions Notes

Abstract

:
Over the years, several machine-learning applications have been suggested to assist in various clinical scenarios relevant to oral cancer. We offer a systematic review to identify, assess, and summarize the evidence for reported uses in the areas of oral cancer detection and prevention, prognosis, pre-cancer, treatment, and quality of life. The main algorithms applied in the context of oral cancer applications corresponded to SVM, ANN, and LR, comprising 87.71% of the total published articles in the field. Genomic, histopathological, image, medical/clinical, spectral, and speech data were used most often to predict the four areas of application found in this review. In conclusion, our study has shown that machine-learning applications are useful for prognosis, diagnosis, and prevention of potentially malignant oral lesions (pre-cancer) and therapy. Nevertheless, we strongly recommended the application of these methods in daily clinical practice.

1. Introduction

Oral cancer has emerged as a serious public health issue across the world. According to the literature, the global incidence, mortality, and disability-adjusted life years of this disease increased by nearly 1.0-fold between 1990 and 2017 [1]. Based on the GLOBOCAN estimates of incidence and mortality, 377,713 new cases and 177,757 deaths for lip and oral cavity cancer were reported for the year 2020 [2]. Most oral cancers are squamous cell carcinomas, which is an aggressive disease with a high tendency to metastasize locally and to distant sites. It has a considerable impact on a patient’s life and on society as a whole. Oral cancer has a 5-year overall survival rate of just approximately 51.7 percent due to frequently late diagnosis [3].
The methods used for oral cancer diagnosis include the traditional anamnesis and clinical examination, complemented with image and hematoxylin–eosin histopathological analysis, the latter being the most common method [4,5]. Immunohistochemistry is routinely used to distinguish the disease in more complex instances and to aid in disease staging. For its examination, molecular approaches have been devised with the goal of finding biomarkers that can anticipate early alterations. In situ hybridization, gel electrophoresis and blotting, flow cytometry, mass spectrometry, polymerase chain reaction, microarrays, Sanger sequencing, and next-generation sequencing are common techniques employed in molecular diagnostics of oral squamous cell carcinomas [6].
In the practice of clinical medicine and in all health-related tasks, the diagnostic process is critical. A correct diagnostic evaluation is essential for the effectiveness of disease therapy. This diagnostic process is based on the interpretation of information supplied by the patient in the anamnesis, as well as the clinician’s clinical examination, in addition to the information provided by the complementary tests. In short, an accurate diagnosis is obtained after analyzing the data of the disease.
Artificial intelligence (AI) is commonly employed among diverse areas of medicine [7,8]. Radiology and sophisticated imaging technologies, pathology, ophthalmology, and dermatology are the disciplines to which it has made significant contributions [7,8]. In each case, a number of impediments must be assessed [8]. Three factors are commonly used to apply the regulations: the danger to patient safety, the presence of a predictive algorithm, and the amount of human involvement [8].
Mathematical models have been used to analyze specific aspects related to oral cancer. Multistage clonal expansion models have been used to analyze the incidence of human papillomavirus (HPV)-related and unrelated oral squamous cell carcinoma, concluding that this model can be useful to identify temporal trends in cancer mechanisms [9]. Additionally, mathematical modeling combined with vitro experimentation has been used to analyze the nanoparticle uptake of oral cancer cells, concluding that the number of receptors per cell was the dominant mechanism in the process [10].
Machine learning (ML) has also been used in oral cancer studies to explore the discrimination between well-differentiated (WD) oral squamous cell carcinoma (OSCC) and moderately or poorly differentiated OSCC [11], to evaluate its ability to predict disease outcome [12], to predict the occurrence of lymph node metastasis of early-stage oral tongue squamous cell carcinoma [13], among other topics of this disease.
ML applications can be classified based on the clinical context of the disease, including diagnosis and prevention, prognosis, potentially malignant oral lesions (pre-cancer), and therapy and quality of life.
According to the NCI Dictionary of Cancer Terms (https://www.cancer.gov/publications/dictionaries/cancer-terms, accessed on 1 March 2022), diagnosis corresponds to the process of identifying a disease, condition, or injury from its signs and symptoms. Cancer prevention includes avoiding risk factors and increasing protective factors. Prognosis is the likely outcome or course of a disease, i.e., the chance of recovery or recurrence. Potentially malignant oral lesions are states of the oral mucosa that are at an increased risk of malignant transformation compared to healthy mucosa [14]. Therapy and quality of life represent the clinical approach to cancer and how it affects an individual’s sense of well-being and ability to carry out activities of daily living.
Considering the current characteristics of oral cancer in terms of the increase in its incidence, the need to strengthen the diagnostic tools, and the existence of abundant literature on the use of machine learning in the study of this disease, a comprehensive study was carried out with the goal of analyzing the potential uses of machine learning in oral cancer.

2. Materials and Methods

This systematic review was conducted according to guidelines reported in the indications of the Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) [15].
This study aimed to answer the question: “Which are the machine-learning applications used in oral cancer?”. For this, a systematic literature search based on keywords was performed. The search was carried out considering the following databases: Web of Science, PubMed, ScienceDirect, and IEEE.

2.1. Search Strategy

The literature search was carried out through four journal databases (Web of Science, PubMed, ScienceDirect, and IEEE) and ran up to 22 May 2020. The search strategy used both medical subject headings (MeSH) and free-text terms. The search with MeSH terms was run through PubMed, and the terms corresponded to: “machine learning” AND “mouth neoplasms”, and “artificial intelligence” AND “mouth neoplasms”. With regard to the free-text terms, they corresponded to: (machine learning AND oral cancer) OR (artificial intelligence AND oral cancer) OR (machine learning AND OSCC) OR (artificial intelligence AND OSCC).

2.2. Inclusion Criteria and Study Selection Process

The studies considered were only studies that dealt with machine-learning applications in the field of oral cancer. Specifically, we sought studies evaluating the different uses of the machine-learning field in oral cancer disease.
For inclusion within this review, studies were selected according to the following inclusion and exclusion criteria:
-
Articles reporting data related to the machine-learning applications in oral cancer disease.
-
Only original articles in English language were considered.
-
Case reports, lectures, data in brief, reviews, in vitro studies (on animals and on human cell lines) and non-original data were excluded from this study.
-
Articles that did not involve a concrete machine-learning application in oral cancer disease were excluded.

2.3. Data Collection and Extraction

In an unblinded but separate approach, two researchers (X.A.L.C. and F.M.) assessed the titles of the papers found by the search method across the four online databases. Articles that were duplicated were removed. The abstracts were then screened by two researchers who worked separately (X.A.L.C. and F.M.). Any article that appeared to fit the inclusion criteria was subjected to a full-text review. Disagreements between the four writers throughout the abstract screening stage and full-text eligibility were settled by consensus.
For data extraction, three researchers (X.A.L.C., C.R., and B.V.) performed a training phase in order to discuss the data extraction (items to consider) from the selected final articles. Finally, two authors were in charge of independently extracting these items (X.A.L.C. and F.M.). This process was cross-checked. All performances were considered when a study reported several classification experiments. In the case when a study compared several feature combinations, the performance of the best combination was considered. Performance analysis was conducted according to the following statistic metrics: accuracy (ACC), sensitivity (SE), specificity (SP), and area under the ROC curve (AUC).
Cohen’s kappa statistic was used to calculate the agreement between the reviewers. In addition, the risk of bias (ROB) in the studies was calculated by using the Prediction model Risk of Bias Assessment Tool (PROBAST) [16]. PROBAST contains a set of 20 signaling questions from four domains, which involve aspects such as participants, predictors, results, and analysis to allow the evaluation of the risk of bias in predictive model studies.

3. Results

The PRISMA flow diagram followed is depicted in Figure 1. In detail, 478 articles were identified in the four databases. After eliminating the duplicate articles and applying the inclusion and exclusion criteria, 63 articles were obtained. From these, six articles were excluded for different reasons, including: out of goal (one article), unavailability online (two articles), and lack of ML techniques (three articles). Finally, 57 articles were included (Figure 1).
The selected studies (57 in total) were screened; the inputs of model, number of samples, outcomes, ML techniques employed, and risk of bias were recorded. These parameters are summarized in Table 1. The first study dates from 1996. Most studies included in this systematic review were published in the years 2018 and 2020.
The included studies were conducted in Brazil, Israel, Taiwan, Japan, Malaysia, Netherlands, and others. The number of samples ranged from 20 to 33,065 (mean 1192). The result of the kappa agreement was 0.85, which classifies as almost perfect. Differences were resolved by consensus of reviewers. Regarding the oral pathologist–oral cancer context, the selected studies were grouped according to different ML applications in the field. These applications correspond to the following clinical contexts: (i) diagnosis and prevention, (ii) prognosis, (iii) potentially malignant oral lesions (pre-cancer), and (iv) therapy and quality of life (Figure 2).
In concordance with the four application areas found, the diagnosis and prevention [19,24,25,29,31,32,33,35,36,38,39,40,41,42,45,46,47,49,56,58,63,65,67,69,70,71] register the largest number of articles with 45.61%, followed by prognosis [17,18,23,26,37,43,48,50,51,52,54,57,62,68,72,73] with 28.07 %, and potentially malignant oral lesions (pre-cancer) [20,21,26,30,34,44,55,59,60,61,64,66] with 21.05%, as shown in Figure 2. On the other hand, 5.26% of the articles focus on therapy and quality of life [22,27,53]. Furthermore, from 2018, a sustained increase in the number of publications that address the prognosis and diagnosis of oral cancer pathology can be observed.
Various machine-learning algorithms were used in the investigations. The first study was conducted in 1996, using the performance of a computer-generated neural network trained to identify normal, premalignant, and malignant oral smears using an artificial neural network [56]. Among the most frequently applied algorithms in oral cancer pathology (Figure 3) is the support vector machine (SVM) with 42.10%, the artificial neural network (ANN) with 24.56%, and the logistic regression (LR) with 21.05%. On the other hand, in the deep-learning subarea, the application of the convolutional neural network (CNN) was 12.28%.
Studies of machine-learning applications in oral cancer analyzed different types of data (Table 1), for example: genomic data [17,19,22,47,48,50,53,54,72], histopathological data [49,56,64], image data [20,21,26,28,29,30,31,32,33,38,39,40,42,44,55,60,61,62,66,67], medical history/clinical data [18,23,25,35,37,43,51,57,59,68,70,71,73], spectral data [24,34,36,41,45,46,52,58,63,65,69], and speech data [27].

3.1. Risk of Bias

In a systematic review, the ROB is a necessary stage. The findings of the ROB assessment were assessed in this fashion using PROBAST, as indicated in Table 1. A ROB judgment was applied to each investigated category in order to determine if the prediction model’s predictive performance/accuracy was likely biased.
Of the total studies analyzed, 84.21% of the studies presented a low ROB, 7.01% presented a high ROB, and 8.77% presented an unclear ROB (Table 1).

3.2. Predictive Model Evaluation

The performance of the methods was reported by the authors in terms of the accuracy, area under the ROC curve, sensitivity, and/or specificity. However, not all of these metrics were reported all the time. In total, 57.89% of the studies reported sensitivity, 63.15% of the studies reported accuracy, 59.64% of the studies reported specificity, and 38.59% of the studies reported AUC. For instance, the metrics of ACC, AUC, SE, and SP, for the most part, used algorithms among the studies (Figure 3) and are shown in Table 2. Specifically, SVM gave an ACC of 85.83%, AUC of 0.83, SE of 86.45%, and SP of 88.20%. ANN gave an ACC of 75.72%, AUC of 0.69, SE of 76.90%, and SP of 84.59%. In the case of LR, the media for ACC, AUC, SE, and SP was 75.47%, 0.7 76.53%, and 77.51%, respectively. In addition, ANOVA analysis was performed (Table 2). Statistical significance was obtained for metrics of ACC, SE, and SP (p-value < 0.05).
According to the studies included in the analysis, 94.73% present some method of validation of the model, as shown in Table 3. Only four studies did not use a validation method (Table 3). Specifically, the “hold out” validation was used in 31.57 % of the studies. The hold out validation consists of a method that divides the dataset into training and test set. The other most used method corresponds to the “cross validation” (CV), with 50.87% of the studies. This validation method considers the division of the set-in k-folds, where it uses one of the subsets as test data and the rest (K-1) as training data. Finally, the most used validations among the different studies were 5 and 10-fold CV (Table 3).
The performance of the top three most frequently applied algorithms (SVM, ANN, and LR) was compared statistically with each other. Algorithms were compared according to ACC, SE, SP, and AUC (Table 2). With respect to ACC performance, the algorithms differed significantly (p < 0.05). The SE performance of SVM, ANN, and LR did not differ significantly (p = 0.058). Regarding SP, the three algorithms did not differ significantly (p = 0.270). Finally, the AUC performance of SVM, ANN, and LR did not differ significantly (p = 0.213) (Table 2).

4. Discussion

In this study, we systematically reviewed the literature and described the state-of-the-art, as well as current, applications of machine learning in oral cancer. In this systematic review, we quantified the chosen studies and classified them according to four areas of application: diagnosis and prevention, prognosis, potentially malignant oral lesions (pre-cancer), and therapy and quality of life. To our knowledge, this is the first systematic review of ML applications in oral cancer.
In recent years, many studies have been published by using genomic, histopathological, image, medical/clinical, spectral, and speech data in combination with machine-learning techniques with the aim of applying this knowledge to the four previously mentioned application areas [17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73].
The most cited machine-learning algorithms in the context of oral cancer applications corresponded to SVM, ANN, and LR, comprising 87.71% of the total published articles in the field. In the area of ANN, it is important to mention the growth of the deep-learning subarea since 2017. This growth can be explained due to the greater availability of data, as well as the greater computing power through architectures that are dedicated to machine learning, and the use of graphic processing units (GPUs) that considerably reduce the processing time of the data [74].
Most of the machine-learning applications were concentrated in the analysis of medical history/clinical data, spectral data, genomic data, and image analysis. With respect to the different clinical contexts analyzed, for diagnosis and prevention, most of the applied algorithms were SVM, ANN, and LR. The best one was SVM due to its AUC value in comparison to ANN and LR. In addition, the ANOVA p-value was 0.039, showing statistical difference between SVM, ANN, and LR (Table 2). In the case of prognosis, most of the applied algorithms were SVM, ANN, and LR. The best one was ANN due to its AUC value (0.76) in comparison to the other algorithms (Table 2). Nevertheless, SVM is also a good predictor due to its homogenous values among the different metrics of ACC, SE, SP, and AUC (0.75). In this way, it is possible to say that SVM and ANN are the most adequate algorithms in the clinical context of prognosis. According to potentially malignant oral lesions (pre-cancer), most of the applied algorithms were also SVM, ANN, and LR. The best one was SVM; the metrics of ACC, SE, and SP were equal to or greater than 89.69% (Table 2). Finally, for therapy and quality of life, most of the applied algorithms were SVM and ANN. The best one was SVM, with and AUC value of 0.89 (Table 2).
The most common machine-learning applications were focused on diagnosis and prevention, followed by prognosis and potentially malignant oral lesions. However, oral cancer is not a particularly difficult malignancy to diagnose. The mouth is a part of the body that is readily accessible for early detection [75]. A great clinical challenge is to establish which lesions will progress to oral cancer. Early diagnosis of oral pre-cancerous lesions is particularly challenging because it requires dentists to be familiar with the range of clinical presentations of potentially malignant oral lesions, many of which may resemble less serious lesions [76]. Here, algorithms have many opportunities to have a clinical impact, for example, in supporting the histopathological evaluation of dysplastic lesions. In this aspect, the applications in potentially malignant oral lesions comprise the 21.05% of the total studies analyzed. Recent studies in this area performed image analysis [21,60,61]. Otherwise, despite the promising results, none of the studies demonstrated an improvement in detecting potentially malignant oral lesions.
According to the most frequent types of data per area of application, we found: (i) spectral data with 40.90% in the diagnosis and prevention area, (ii) medical/clinical data with 50.00% in prognosis, (iii) image data with 72.72% in potentially malignant oral lesions (pre-cancer), and (iv) genomic data representing 66.66% in the therapy and quality of life application area.
As mentioned before, deep-learning has been applied in diverse areas (Table 1). In specific, it was used for the analysis of image and spectral data (Table 1) [20,21,24,26,28,29,30,31,32,33,34,36,38,39,40,41,42,44,45,46,52,55,58,60,61,62,63,65,66,67,69].
The main areas of application of deep-learning were in diagnosis and prevention, and potentially malignant oral lesions (pre-cancer). In both areas, the images were the most used type of data, representing 83.33% and 100% of the total studies for diagnosis and prevention, and potentially malignant oral lesions (pre-cancer), respectively. The use of new technologies in high-quality image acquisition devices has made it possible to develop and enhance deep-learning approaches by employing convolutional neural networks.
SVM is among the most used algorithms in ML by area of application. SVM is the focus of the largest number of studies, representing 45.45% and 63.63% of the total studies for diagnosis and prevention, and potentially malignant oral lesions (precancer), respectively. Furthermore, for prognosis, LR was the most frequently used algorithm, with 43.75%, while in therapy and quality of life, the algorithms CTREE, RF, BA, SVM, ANN, and DFG represented 100%. On the other hand, in the deep-learning subarea, the CNN stood out in diagnosis and prevention, and in potentially malignant oral lesions (precancer) with 100%.
Studies varied according to the algorithm applied, the input and output variables, and the methods for assessing predictive model performance. SVM, ANN, and LR were the most commonly applied algorithms. Genomic, histopathological, image, medical/clinical, spectral, and speech data were the most often used to predict the four areas of application found in this review.
ML, with algorithms such as SVM, ANN, LR, CNN, represents a powerful method, capable of effectively predicting outcomes in order to support diagnosis and prevention, prognosis, potentially malignant oral lesions (pre-cancer), and therapy and quality of life.
It is important to note that not all these algorithms are intuitive. For example, ANN and SVM are nonlinear and inscrutable in the way they generate their outputs. In this sense, clinicians tend to lack trust in the outputs of a clinical decision support system when it is not clear how the algorithm gives the classification result, unlike decision trees, which identify the set of rules that there are behind the classification. In this aspect, it is very important to promote the transparency of these methods, which can be useful to facilitate their implementation in the field. This transparency can be achieved by reporting, in the case of SVM, the values of sigma (available only in some kernel types), C parameter, and type of kernel, whereas for ANN, it is important to report the number of layers and the corresponding number of nodes or neurons.
Of the total included articles, most of the studies showed a low risk of bias. In addition, most of the studies used some type of validation. The most commonly applied validation method corresponded to a 5-fold CV and a 10-fold CV. In this aspect, avoiding the overfitting was a task considered in most of the studies included. Nevertheless, not all of the performance metrics were reported in all of the studies. Future studies would do well to report at least AUC, ACC, SE, and SP due to their importance in the analysis of the machine-learning method in order to enable the comparison between studies and facilitate performance evaluation.
Additionally, not all the metrics were reported all the time; 63.15% of the studies reported accuracy, 29.82% of the studies did not report sensitivity, 38.59% of the studies did not report specificity, and 59.64 % of the studies did not report AUC.
Today, there are no tumor biomarkers routinely used in the clinical setting to predict high-risk oral dysplastic lesions [77]. The recognition of histological patterns that may go unnoticed by the pathologist could provide an opportunity for early therapeutic interventions, which would improve the prognosis. It is known that early diagnosis of oral cancer is associated with high survival rates. Therefore, if carcinogenesis is seen like an arrow from left to right, being placed leftmost of the passage of a susceptibility state to a canzerisable field is even better [78]. Unfortunately, our results show that the frequency of studies in this field is still low.
One of the findings in this systematic review is that diagnosis and prevention is the main field approached in the articles analyzed. This is coincident with what we find in the clinical practice of oral pathology and medicine. Most oral cancer diagnoses are made in advanced stages of the disease, i.e., in the III and IV TNM stages. The result is the existence of poor survival rates, plus a severe morbidity in survivors due to strong sequelae associated with surgery and radiotherapy. It is interesting to note that, in spite of the great advances in oncology research, which has brought considerable improvements in oncotherapy, the diagnostic problem continues to be unsolved. Great advances have also been made in early diagnosis, mainly based on general practitioners training, but the problem persists in most countries. Two approaches that look for solutions can be considered: prevention strategies for patients and new additional methods to assist the diagnostic process. According to this systematic review, machine learning could be among these new methods.
In addition to the diagnostics issue, the prognostic approach is also a problem that must be analyzed. When a clinician finds a white lesion in the oral mucosa, they have to decide about its possible cancerization. Leukoplakias vary in their percentages of transformation to malignant states. Biopsy will help that decision, but the clinician and the pathologist also frequently need additional information. Neural networks have been proved a useful method for discrimination between leukoplakia and normal tissue by using tissue autofluorescence spectra properties. Although it must be considered as a complementary tool after clinical examination, it helps determine the intrinsic abnormalities of oral mucosa that can lead to a malignant disease.
From a therapeutic point of view, usually, the surgical team has to decide if neck dissection needs to be performed or not, following primary tumor excision. Depth of invasion (DOI) has been shown to be useful to predict nodal metastasis, but it is still frequent to find N0 patients with neck recurrences after surgery without neck dissection, and, on the other hand, neck dissections with no nodal metastasis are found when microscopic examination is conducted. This systematic review found that machine-learning algorithms can be useful when this decision has to be made.

5. Conclusions

To conclude, our study showed that machine-learning applications can be useful in different areas of the oral cancer disease. These areas include prognosis, diagnosis and prevention, potentially malignant oral lesions (pre-cancer), and therapy.
Regarding the most suitable algorithms by area of application in oral cancer, we can say that the SVM is more appropriate for the diagnosis and prevention clinical context, and the ANN and SVM are the most suitable for the prognosis clinical context in terms of the performance of the algorithms. For potentially malignant oral lesions (pre-cancer) and therapy and quality of life, the data do not allow us to determine which is the most appropriate algorithm due to the smaller number of studies. Nevertheless, despite the few data reported, we can suggest that SVM and ANN are potentially appropriate algorithms to apply in the clinical context of pre-cancer and therapy.
We strongly suggest continuing exploring the application of these new methods in daily clinical practice.

Author Contributions

Conceptualization, X.A.L.-C. and C.R.; methodology, X.A.L.-C., B.V. and C.R.; software, X.A.L.-C. and F.M.; validation, X.A.L.-C., B.V., C.R. and F.M.; formal analysis, X.A.L.-C., B.V., C.R. and F.M.; investigation, X.A.L.-C., B.V., C.R. and F.M.; resources, X.A.L.-C. and F.M.; data curation, X.A.L.-C. and F.M.; writing—original draft preparation, X.A.L.-C., B.V., C.R. and F.M.; writing—review and editing, X.A.L.-C., B.V., C.R. and F.M.; visualization, X.A.L.-C. and F.M.; supervision, X.A.L.-C., B.V. and C.R.; project administration, X.A.L.-C.; funding acquisition, X.A.L.-C. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the Program of International Cooperation from the Agencia Nacional de Investigación y Desarrollo (ANID), grant number REDI170172.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

Not applicable.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Ren, Z.; Hu, C.; He, H.; Li, Y.; Lyu, J. Global and regional burdens of oral cancer from 1990 to 2017: Results from the global burden of disease study. Cancer Commun. 2020, 40, 81–92. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  2. Lip, Oral Cavity Fact Sheet. 2020. Available online: https://gco.iarc.fr/today/data/factsheets/cancers/1-Lip-oral-cavity-fact-sheet.pdf (accessed on 1 March 2022).
  3. Kowalski, L.P.; de Oliveira, M.M.; Lopez, R.V.M.; e Silva, D.R.M.; Ikeda, M.K.; Curado, M.P. Survival trends of patients with oral and oropharyngeal cancer treated at a cancer center in São Paulo, Brazil. Clinics 2020, 75, e1507. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  4. Roi, A.; Roi, C.I.; Andreescu, N.I.; Riviş, M.; Badea, I.D.; Meszaros, N.; Rusu, L.C.; Iurciuc, S. Oral cancer histopathological subtypes in association with risk factors: A 5-year retrospective study. Rom. J. Morphol. Embryol. 2020, 61, 1213–1220. [Google Scholar] [CrossRef] [PubMed]
  5. Leite, C.F.; Silva, K.D.D.; Horta, M.C.R.; de Aguiar, M.C.F. Can morphological features evaluated in oral cancer biopsies influence in decision-making? A preliminary study. Pathol. Res. Pract. 2020, 216, 153138. [Google Scholar] [CrossRef]
  6. Yap, T.; Celentano, A.; Seers, C.; McCullough, M.J.; Farah, C.S. Molecular diagnostics in oral cancer and oral potentially malignant disorders—A clinician’s guide. J. Oral Pathol. Med. 2020, 49, 1–8. [Google Scholar] [CrossRef]
  7. Tapia-Castillo, A.; Carvajal, C.A.; López-Cortés, X.; Vecchiola, A.; Fardella, C.E. Novel metabolomic profile of subjects with non-classic apparent mineralocorticoid excess. Sci. Rep. 2021, 11, 17156. [Google Scholar] [CrossRef]
  8. Kulkarni, S.; Seneviratne, N.; Baig, M.S.; Khan, A.H.A. Artificial intelligence in medicine: Where are we now? Acad. Radiol. 2020, 27, 62–70. [Google Scholar] [CrossRef] [Green Version]
  9. Brouwer, A.F.; Eisenberg, M.C.; Meza, R. Age effects and temporal trends in HPV-related and HPV-unrelated oral cancer in the United States: A multistage carcinogenesis modeling analysis. PLoS ONE 2016, 11, e0151098. [Google Scholar] [CrossRef]
  10. Sorrell, I.; Shipley, R.J.; Hearnden, V.; Colley, H.E.; Thornhill, M.H.; Murdoch, C.; Webb, S.D. Combined mathematical modelling and experimentation to predict polymersome uptake by oral cancer cells. Nanomed. Nanotechnol. Biol. Med. 2014, 10, 339–348. [Google Scholar] [CrossRef] [Green Version]
  11. Ren, J.; Qi, M.; Yuan, Y.; Duan, S.; Tao, X. Machine learning–based MRI texture analysis to predict the histologic grade of oral squamous cell carcinoma. Am. J. Roentgenol. 2020, 215, 1184–1190. [Google Scholar] [CrossRef]
  12. Chu, C.S.; Lee, N.P.; Adeoye, J.; Thomson, P.; Choi, S. Machine learning and treatment outcome prediction for oral cancer. J. Oral Pathol. Med. 2020, 49, 977–985. [Google Scholar] [CrossRef]
  13. Shan, J.; Jiang, R.; Chen, X.; Zhong, Y.; Zhang, W.; Xie, L.; Cheng, J.; Jiang, H. Machine learning predicts lymph node metastasis in early-stage oral tongue squamous cell carcinoma. J. Oral Maxillofac. Surg. 2020, 78, 2208–2218. [Google Scholar] [CrossRef]
  14. Hadzic, S.; Gojkov-Vukelic, M.; Pasic, E.; Dervisevic, A. Importance of early detection of potentially malignant lesions in the prevention of oral cancer. Mater. Socio-Med. 2017, 29, 129–133. [Google Scholar] [CrossRef] [Green Version]
  15. Moher, D.; Liberati, A.; Tetzlaff, J.; Altman, D.G. Preferred reporting items for systematic reviews and meta-analyses: The PRISMA statement. BMJ 2009, 339, b2535. [Google Scholar] [CrossRef] [Green Version]
  16. Wolff, R.F.; Moons, K.G.M.; Riley, R.D.; Whiting, P.F.; Westwood, M.; Collins, G.S.; Reitsma, J.B.; Kleijnen, J.; Mallett, S. PROBAST: A tool to assess the risk of bias and applicability of prediction model studies. Ann. Intern. Med. 2019, 170, 51. [Google Scholar] [CrossRef] [Green Version]
  17. Cao, R.; Wu, Q.; Li, Q.; Yao, M.; Zhou, H. A 3-mRNA-based prognostic signature of survival in oral squamous cell carcinoma. PeerJ 2019, 7, e7360. [Google Scholar] [CrossRef] [Green Version]
  18. Tan, M.S.; Tan, J.W.; Chang, S.-W.; Yap, H.J.; Abdul Kareem, S.; Zain, R.B. A genetic programming approach to oral cancer prognosis. PeerJ 2016, 4, e2482. [Google Scholar] [CrossRef] [Green Version]
  19. Hsieh, C.-H.; Chen, W.-M.; Hsieh, Y.-S.; Fan, Y.-C.; Yang, P.E.; Kang, S.-T.; Liao, C.-T. A novel multi-gene detection platform for the analysis of miRNA expression. Sci. Rep. 2018, 8, 10684. [Google Scholar] [CrossRef] [Green Version]
  20. Paul, R.R. A novel wavelet neural network based pathological stage detection technique for an oral precancerous condition. J. Clin. Pathol. 2005, 58, 932–938. [Google Scholar] [CrossRef] [Green Version]
  21. Sunny, S.; Baby, A.; James, B.L.; Balaji, D.; Aparna, N.V.; Rana, M.H.; Gurpur, P.; Skandarajah, A.; D’Ambrosio, M.; Ramanjinappa, R.D.; et al. A smart tele-cytology point-of-care platform for oral cancer screening. PLoS ONE 2019, 14, e0224885. [Google Scholar] [CrossRef] [Green Version]
  22. Randhawa, V.; Kumar Singh, A.; Acharya, V. A systematic approach to prioritize drug targets using machine learning, a molecular descriptor-based classification model, and high-throughput screening of plant derived molecules: A case study in oral cancer. Mol. Biosyst. 2015, 11, 3362–3377. [Google Scholar] [CrossRef]
  23. Cheng, C.S.; Shueng, P.W.; Chang, C.C.; Kuo, C.W. Adapting an evidence-based diagnostic model for predicting recurrence risk factors of oral cancer. J. Univ. Comput. Sci. 2018, 24, 742–752. [Google Scholar]
  24. Kan, C.; Lee, A.Y.; Nieman, L.T.; Sokolov, K.; Markey, M.K. Adaptive spectral window sizes for extraction of diagnostic features from optical spectra. J. Biomed. Opt. 2010, 15, 047012. [Google Scholar] [CrossRef]
  25. Downer, M.C.; Jullien, J.A.; Speight, P.M. An interim determination of health gain from oral cancer and precancer screening: 3. Preselecting high risk individuals. Community Dent. Health 1998, 15, 72–76. [Google Scholar] [PubMed]
  26. Lu, C.; Lewis, J.S.; Dupont, W.D.; Plummer, W.D.; Janowczyk, A.; Madabhushi, A. An oral cavity squamous cell carcinoma quantitative histomorphometric-based image classifier of nuclear morphology can risk stratify patients for disease-specific survival. Mod. Pathol. 2017, 30, 1655–1665. [Google Scholar] [CrossRef] [PubMed]
  27. de Bruijn, M.; ten Bosch, L.; Kuik, D.J.; Langendijk, J.A.; Leemans, C.R.; Leeuw, I.V. Artificial neural network analysis to assess hypernasality in patients treated for oral or oropharyngeal cancer. Logop. Phoniatr. Vocol. 2011, 36, 168–174. [Google Scholar] [CrossRef] [PubMed]
  28. Muthu Rama Krishnan, M.; Pal, M.; Bomminayuni, S.K.; Chakraborty, C.; Paul, R.R.; Chatterjee, J.; Ray, A.K. Automated classification of cells in sub-epithelial connective tissue of oral sub-mucous fibrosis—An SVM based approach. Comput. Biol. Med. 2009, 39, 1096–1104. [Google Scholar] [CrossRef]
  29. Das, N.; Hussain, E.; Mahanta, L.B. Automated classification of cells into multiple classes in epithelial tissue of oral squamous cell carcinoma using transfer learning and convolutional neural network. Neural Netw. 2020, 128, 47–60. [Google Scholar] [CrossRef]
  30. Krishnan, M.M.R.; Acharya, U.R.; Chakraborty, C.; Ray, A.K. Automated diagnosis of oral cancer using higher order spectra features and local binary pattern: A comparative study. Technol. Cancer Res. Treat. 2011, 10, 443–455. [Google Scholar] [CrossRef]
  31. Rahman, T.Y.; Mahanta, L.B.; Das, A.K.; Sarma, J.D. Automated oral squamous cell carcinoma identification using shape, texture and color features of whole image strips. Tissue Cell 2020, 63, 101322. [Google Scholar] [CrossRef]
  32. Aubreville, M.; Knipfer, C.; Oetter, N.; Jaremenko, C.; Rodner, E.; Denzler, J.; Bohr, C.; Neumann, H.; Stelzle, F.; Maier, A. Automatic classification of cancerous tissue in laserendomicroscopy images of the oral cavity using deep learning. Sci. Rep. 2017, 7, 11979. [Google Scholar] [CrossRef] [Green Version]
  33. Das, D.K.; Bose, S.; Maiti, A.K.; Mitra, B.; Mukherjee, G.; Dutta, P.K. Automatic identification of clinically relevant regions from oral tissue histological images for oral squamous cell carcinoma diagnosis. Tissue Cell 2018, 53, 111–119. [Google Scholar] [CrossRef]
  34. Van Staveren, H.J.; Van Veen, R.L.P.; Speelman, O.C.; Witjes, M.J.H.; Star, W.M.; Roodenburg, J.L.N. Classification of clinical autofluorescence spectra of oral leukoplakia using an artificial neural network: A pilot study. Oral Oncol. 2000, 36, 286–293. [Google Scholar] [CrossRef]
  35. Ge, S.; Lu, H.; Li, Q.; Logan, H.L.; Dodd, V.J.; Bian, J.; Shenkman, E.A.; Guo, Y. Classification tree analysis of factors associated with oral cancer exam. Am. J. Health Behav. 2019, 43, 635–647. [Google Scholar] [CrossRef]
  36. de Veld, D.C.G.; Skurichina, M.; Witjes, M.J.H.; Duin, R.P.W.; Sterenborg, H.J.C.M.; Roodenburg, J.L.N. Clinical study for classification of benign, dysplastic, and malignant oral lesions using autofluorescence spectroscopy. J. Biomed. Opt. 2004, 9, 940–950. [Google Scholar] [CrossRef] [Green Version]
  37. Schwarzer, G.; Nagata, T.; Mattern, D.; Schmelzeisen, R.; Schumacher, M. Comparison of fuzzy inference, logistic regression, and classification trees (CART): Prediction of cervical lymph node metastasis in carcinoma of the tongue. Methods Inf. Med. 2003, 42, 572–577. [Google Scholar] [CrossRef]
  38. Das, D.K.; Mitra, P.; Chakraborty, C.; Chatterjee, S.; Maiti, A.K.; Bose, S. Computational approach for mitotic cell detection and its application in oral squamous cell carcinoma. Multidimens. Syst. Signal Process. 2017, 28, 1031–1050. [Google Scholar] [CrossRef]
  39. Jeyaraj, P.R.; Samuel Nadar, E.R. Computer-assisted medical image classification for early diagnosis of oral cancer employing deep learning algorithm. J. Cancer Res. Clin. Oncol. 2019, 145, 829–837. [Google Scholar] [CrossRef]
  40. Ariji, Y.; Fukuda, M.; Kise, Y.; Nozawa, M.; Yanashita, Y.; Fujita, H.; Katsumata, A.; Ariji, E. Contrast-enhanced computed tomography image assessment of cervical lymph node metastasis in patients with oral cancer by using a deep learning system of artificial intelligence. Oral Surg. Oral Med. Oral Pathol. Oral Radiol. 2019, 127, 458–463. [Google Scholar] [CrossRef]
  41. Yu, M.; Yan, H.; Xia, J.; Zhu, L.; Zhang, T.; Zhu, Z.; Lou, X.; Sun, G.; Dong, M. Deep convolutional neural networks for tongue squamous cell carcinoma classification using Raman spectroscopy. Photodiagnosis Photodyn. Ther. 2019, 26, 430–435. [Google Scholar] [CrossRef]
  42. Dong, F.; Tao, C.; Wu, J.; Su, Y.; Wang, Y.; Wang, Y.; Guo, C.; Lyu, P. Detection of cervical lymph node metastasis from oral cavity cancer using a non-radiating, noninvasive digital infrared thermal imaging system. Sci. Rep. 2018, 8, 7219. [Google Scholar] [CrossRef]
  43. Karadaghy, O.A.; Shew, M.; New, J.; Bur, A.M. Development and assessment of a machine learning model to help predict survival among patients with oral squamous cell carcinoma. JAMA Otolaryngol. Neck Surg. 2019, 145, 1115–1120. [Google Scholar] [CrossRef]
  44. Spyridonos, P.; Gaitanis, G.; Bassukas, I.D.; Tzaphlidou, M. Evaluation of vermillion border descriptors and relevance vector machines discrimination model for making probabilistic predictions of solar cheilosis on digital lip photographs. Comput. Biol. Med. 2015, 63, 11–18. [Google Scholar] [CrossRef]
  45. Banerjee, S.; Pal, M.; Chakrabarty, J.; Petibois, C.; Paul, R.R.; Giri, A.; Chatterjee, J. Fourier-transform-infrared-spectroscopy based spectral-biomarker selection towards optimum diagnostic differentiation of oral leukoplakia and cancer. Anal. Bioanal. Chem. 2015, 407, 7935–7943. [Google Scholar] [CrossRef]
  46. Zlotogorski-Hurvitz, A.; Dekel, B.Z.; Malonek, D.; Yahalom, R.; Vered, M. FTIR-based spectrum of salivary exosomes coupled with computational-aided discriminating analysis in the diagnosis of oral cancer. J. Cancer Res. Clin. Oncol. 2019, 145, 685–694. [Google Scholar] [CrossRef] [PubMed]
  47. Yang, Z.; Liang, X.; Fu, Y.; Liu, Y.; Zheng, L.; Liu, F.; Li, T.; Yin, X.; Qiao, X.; Xu, X. Identification of AUNIP as a candidate diagnostic and prognostic biomarker for oral squamous cell carcinoma. EBioMedicine 2019, 47, 44–57. [Google Scholar] [CrossRef] [Green Version]
  48. Winck, F.V.; Prado Ribeiro, A.C.; Ramos Domingues, R.; Ling, L.Y.; Riaño-Pachón, D.M.; Rivera, C.; Brandão, T.B.; Gouvea, A.F.; Santos-Silva, A.R.; Coletta, R.D.; et al. Insights into immune responses in oral cancer through proteomic analysis of saliva and salivary extracellular vesicles. Sci. Rep. 2015, 5, 16305. [Google Scholar] [CrossRef]
  49. Hsu, C.-W.; Chen, Y.-T.; Hsieh, Y.-J.; Chang, K.-P.; Hsueh, P.-C.; Chen, T.-W.; Yu, J.-S.; Chang, Y.-S.; Li, L.; Wu, C.-C. Integrated analyses utilizing metabolomics and transcriptomics reveal perturbation of the polyamine pathway in oral cavity squamous cell carcinoma. Anal. Chim. Acta 2019, 1050, 113–122. [Google Scholar] [CrossRef] [PubMed]
  50. Randhawa, V.; Acharya, V. Integrated network analysis and logistic regression modeling identify stage-specific genes in oral squamous cell carcinoma. BMC Med. Genom. 2015, 8, 39. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  51. Bur, A.M.; Holcomb, A.; Goodwin, S.; Woodroof, J.; Karadaghy, O.; Shnayder, Y.; Kakarala, K.; Brant, J.; Shew, M. Machine learning to predict occult nodal metastasis in early oral squamous cell carcinoma. Oral Oncol. 2019, 92, 20–25. [Google Scholar] [CrossRef] [PubMed]
  52. Paul, A.; Srivastava, S.; Roy, R.; Anand, A.; Gaurav, K.; Husain, N.; Jain, S.; Sonkar, A.A. Malignancy prediction among tissues from oral SCC patients including neck invasions: A 1H HRMAS NMR based metabolomic study. Metab. Off. J. Metab. Soc. 2020, 16, 38. [Google Scholar] [CrossRef] [PubMed]
  53. Vittal, S.; Karthikeyan, G. Modeling association detection in order to discover compounds to inhibit oral cancer. J. Biomed. Inform. 2018, 84, 159–163. [Google Scholar] [CrossRef] [PubMed]
  54. Chang, S.-W.; Abdul-Kareem, S.; Merican, A.F.; Zain, R.B. Oral cancer prognosis based on clinicopathologic and genomic markers using a hybrid of feature selection and machine learning methods. BMC Bioinform. 2013, 14, 170. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  55. Mukherjee, A.; Paul, R.R.; Chaudhuri, K.; Chatterjee, J.; Pal, M.; Banerjee, P.; Mukherjee, K.; Banerjee, S.; Dutta, P.K. Performance analysis of different wavelet feature vectors in quantification of oral precancerous condition. Oral Oncol. 2006, 42, 914–928. [Google Scholar] [CrossRef]
  56. Brickley, M.R.; Cowpe, J.G.; Shepherd, J.P. Performance of a computer simulated neural network trained to categorise normal, premalignant and malignant oral smears. J. Oral Pathol. Med. 1996, 25, 424–428. [Google Scholar] [CrossRef]
  57. Campisi, G.; Calvino, F.; Carinci, F.; Matranga, D.; Carella, M.; Mazzotta, M.; Rubini, C.; Panzarella, V.; Santarelli, A.; Fedele, S.; et al. Peri-tumoral inflammatory cell infiltration in OSCC: A reliable marker of local recurrence and prognosis? An investigation using artificial neural networks. Int. J. Immunopathol. Pharmacol. 2011, 24, 113–120. [Google Scholar] [CrossRef]
  58. Wang, C.-Y.; Tsai, T.; Chen, H.-M.; Chen, C.-T.; Chiang, C.-P. PLS-ANN based classification model for oral submucous fibrosis and oral carcinogenesis. Lasers Surg. Med. 2003, 32, 318–326. [Google Scholar] [CrossRef]
  59. McRae, M.P.; Modak, S.S.; Simmons, G.W.; Trochesset, D.A.; Kerr, A.R.; Thornhill, M.H.; Redding, S.W.; Vigneswaran, N.; Kang, S.K.; Christodoulides, N.J.; et al. Point-of-care oral cytology tool for the screening and assessment of potentially malignant oral lesions. Cancer Cytopathol. 2020, 128, 207–220. [Google Scholar] [CrossRef] [Green Version]
  60. Uthoff, R.D.; Song, B.; Sunny, S.; Patrick, S.; Suresh, A.; Kolur, T.; Keerthi, G.; Spires, O.; Anbarani, A.; Wilder-Smith, P.; et al. Point-of-care, smartphone-based, dual-modality, dual-view, oral cancer screening device with neural network classification for low-resource communities. PLoS ONE 2018, 13, e0207493. [Google Scholar] [CrossRef]
  61. Dey, S.; Sarkar, R.; Chatterjee, K.; Datta, P.; Barui, A.; Maity, S.P. Pre-cancer risk assessment in habitual smokers from DIC images of oral exfoliative cells using active contour and SVM analysis. Tissue Cell 2017, 49, 296–306. [Google Scholar] [CrossRef]
  62. Romeo, V.; Cuocolo, R.; Ricciardi, C.; Ugga, L.; Cocozza, S.; Verde, F.; Stanzione, A.; Napolitano, V.; Russo, D.; Improta, G.; et al. Prediction of tumor grade and nodal status in oropharyngeal and oral cavity squamous-cell carcinoma using a radiomic approach. Anticancer Res. 2020, 40, 271–280. [Google Scholar] [CrossRef]
  63. Nayak, G.S.; Kamath, S.; Pai, K.M.; Sarkar, A.; Ray, S.; Kurien, J.; D’Almeida, L.; Krishnanand, B.R.; Santhosh, C.; Kartha, V.B.; et al. Principal component analysis and artificial neural network analysis of oral tissue fluorescence spectra: Classification of normal premalignant and malignant pathological conditions. Biopolymers 2006, 82, 152–166. [Google Scholar] [CrossRef]
  64. Liu, Y.; Li, Y.; Fu, Y.; Liu, T.; Liu, X.; Zhang, X.; Fu, J.; Guan, X.; Chen, T.; Chen, X.; et al. Quantitative prediction of oral cancer risk in patients with oral leukoplakia. Oncotarget 2017, 8, 46057–46064. [Google Scholar] [CrossRef] [Green Version]
  65. Majumder, S.K.; Ghosh, N.; Gupta, P.K. Relevance vector machine for optical diagnosis of cancer. Lasers Surg. Med. 2005, 36, 323–333. [Google Scholar] [CrossRef]
  66. Muthu Rama Krishnan, M.; Shah, P.; Chakraborty, C.; Ray, A.K. Statistical analysis of textural features for improved classification of oral histopathological images. J. Med. Syst. 2012, 36, 865–881. [Google Scholar] [CrossRef]
  67. Nawandhar, A.; Kumar, N.; Veena, R.; Yamujala, L. Stratified squamous epithelial biopsy image classifier using machine learning and neighborhood feature selection. Biomed. Signal Process. Control 2020, 55, 101671. [Google Scholar] [CrossRef]
  68. Tseng, W.-T.; Chiang, W.-F.; Liu, S.-Y.; Roan, J.; Lin, C.-N. The application of data mining techniques to oral cancer prognosis. J. Med. Syst. 2015, 39, 59. [Google Scholar] [CrossRef]
  69. Brouwer de Koning, S.G.; Baltussen, E.J.M.; Karakullukcu, M.B.; Dashtbozorg, B.; Smit, L.A.; Dirven, R.; Hendriks, B.H.W.; Sterenborg, H.J.C.M.; Ruers, T.J.M. Toward complete oral cavity cancer resection using a handheld diffuse reflectance spectroscopy probe. J. Biomed. Opt. 2018, 23, 1–8. [Google Scholar] [CrossRef]
  70. Sharma, N.; Om, H. Usage of probabilistic and general regression neural network for early detection and prevention of oral cancer. Sci. World J. 2015, 2015, 234191. [Google Scholar] [CrossRef] [Green Version]
  71. Campisi, G.; Di Fede, O.; Giovannelli, L.; Capra, G.; Greco, I.; Calvino, F.; Maria Florena, A.; Lo Muzio, L. Use of fuzzy neural networks in modeling relationships of HPV infection with apoptotic and proliferation markers in potentially malignant oral lesions. Oral Oncol. 2005, 41, 994–1004. [Google Scholar] [CrossRef]
  72. Carnielli, C.M.; Macedo, C.C.S.; De Rossi, T.; Granato, D.C.; Rivera, C.; Domingues, R.R.; Pauletti, B.A.; Yokoo, S.; Heberle, H.; Busso-Lopes, A.F.; et al. Combining discovery and targeted proteomics reveals a prognostic signature in oral cancer. Nat. Commun. 2018, 9, 3598. [Google Scholar] [CrossRef]
  73. Muzio, L.L.; D’Angelo, M.; Procaccini, M.; Bambini, F.; Calvino, F.; Florena, A.M.; Franco, V.; Giovannelli, L.; Ammatuna, P.; Campisi, G. Expression of cell cycle markers and human papillomavirus infection in oral squamous cell carcinoma: Use of fuzzy neural networks. Int. J. Cancer 2005, 115, 717–723. [Google Scholar] [CrossRef]
  74. Steinkraus, D.; Buck, I.; Simard, P.Y. Using GPUs for machine learning algorithms. In Proceedings of the Eighth International Conference on Document Analysis and Recognition (ICDAR’05), Seoul, Korea, 31 August–1 September 2005; Volume 2, pp. 1115–1120. [Google Scholar]
  75. Jafari, A.; Najafi, S.; Moradi, F.; Kharazifard, M.; Khami, M. Delay in the diagnosis and treatment of oral cancer. J. Dent. 2013, 14, 146–150. [Google Scholar]
  76. Ford, P.J.; Farah, C.S. Early detection and diagnosis of oral cancer: Strategies for improvement. J. Cancer Policy 2013, 1, e2–e7. [Google Scholar] [CrossRef] [Green Version]
  77. Rivera, C.; Gallegos, R.; Figueroa, C. Biomarkers of progression to oral cancer in patients with dysplasia: A systematic review. Mol. Clin. Oncol. 2020, 13, 42. [Google Scholar] [CrossRef]
  78. Rivera, C. The challenge of the state of susceptibility to oral cancer. J. Oral Res. 2015, 4, 8–9. [Google Scholar] [CrossRef]
Applsci 12 05715 g001 550
Figure 1. Flow diagram of literature search and selection criteria.
Figure 1. Flow diagram of literature search and selection criteria.
Applsci 12 05715 g001
Applsci 12 05715 g002 550
Figure 2. Year-wise distribution of the number of publications on oral cancer context.
Figure 2. Year-wise distribution of the number of publications on oral cancer context.
Applsci 12 05715 g002
Applsci 12 05715 g003 550
Figure 3. Distribution of publications per year according to different algorithms employed. In this graphic are plotted those algorithms that are most usually used.
Figure 3. Distribution of publications per year according to different algorithms employed. In this graphic are plotted those algorithms that are most usually used.
Applsci 12 05715 g003
Table
Table 1. Original research studies that applied machine-learning methods to oral cancer pathology.
Table 1. Original research studies that applied machine-learning methods to oral cancer pathology.
Ref.YearClinical
Context		Applied
Algorithm		Input
Features		n
Samples		ROBConcluding
Remarks
[17]2019PrognosisXGBOOSTExpression profiles and clinical data291+A three-mRNA signature (CLEC3B, C6, and CLCN1) successfully predicted the survival of OSCC patients
[18]2016PrognosisGP, SVM, LRPersonal details, medical history, p53, p6331+Genetic programing (GP) an ideal prediction model for cancer clinical and genomic data
[19]2018Diagnosis and PreventionSVMmiRNA expression122+Using the platform with an ML algorithm, it discovers miRNA expression patterns capable of separating healthy subjects from OSCC patients
[20]2005Potentially malignant oral lesions (pre-cancer)WNNTEM images of collagen fibers from oral subepithelial region145+The trained network was able to classify normal and oral pre-cancer stages
[21]2019Potentially malignant oral lesions (pre-cancer)SVM, RF, LR, LDA, KNNCytology images60+Applicability of tele-cytology for accurate, remote diagnosis and use of automated ANN-based analysis in improving its efficacy
[22] 2015Therapy and quality of lifeCTREE, RF, BA, SVMGene expression data486+Analyzed the dysregulated gene pairs between control and tumor samples and then implemented an ensemble-based feature selection approach to prioritize targets in oral squamous cell carcinoma (OSCC) for therapeutic exploration
[23]2018PrognosisKSTAR, IBK, RFC, RTPersonal details, medical history, smoking, betel nut chewing, and drinking1428+Evidence-based diagnostic model using machine-learning techniques for the prediction of risk factors of recurrent oral cancer
[24]2010Diagnosis and PreventionLDASpectral data57+Presents an approach to adaptively adjust spectral window sizes for feature extraction from optical spectra
[25]1998Diagnosis and PreventionANNPersonal details, dental attendance, and smoking and drinking habits2027+Sensitivity analysis using a decision model to simulate opportunistic screening for oral cancer and pre-cancer
[26]2017PrognosisLDA, QDA, RF, SVMImages of H&E-stained tissue sections115+Investigates computer-extracted image features of nuclear shape and texture on digitized images of H&E-stained tissue sections for risk stratification of oral cavity squamous cell carcinoma patients compared with standard clinical and pathologic parameters
[27]2011Therapy and quality of lifeANNSpeech recording51-Applicability of neural network feature analysis of nasalance in speech to assess hypernasality in speech of patients treated for oral or oropharyngeal cancer
[28]2009Potentially malignant oral lesions (pre-cancer)SVMImages of SECT (sub-epithelial connective tissue) of NOM and OS20+Automated classification method using SVM for understanding the deviation of normal structural profile of oral mucosa during precancerous changes
[29]2020Diagnosis and PreventionCNNHistopathological images8321+CNN-based multi-class grading method of OSCC could be used for diagnosis of patients with OSCC
[30]2011Potentially malignant oral lesions (pre-cancer)SVMImages of surface epithelium from oral mucosa158+Classification based on textural features for the development of a computer-assisted screening of oral sub-mucous fibrosis (OSF)
[31]2020Diagnosis and PreventionDTREE, SVM, KNN, LDA, LRHistopathological images720+SVM and linear discriminant classifier gave the best result for texture and color features, respectively, from the histopathological images
[32]2017Diagnosis and PreventionCNNImages of confocal laser endomicroscopy (CLE) 7894+Novel automatic approach for OSCC diagnosis using deep-learning technologies on CLE images
[33]2018Diagnosis and PreventionCNN, RFMicroscopic images of the oral mucosa100+CNN approach is proposed for segmentation of different constituent layers from oral mucosa histology images
[34]2000Potentially malignant oral lesions (pre-cancer)ANNSpectral data28-Neural networks provide a very good discrimination between autofluorescence spectra of leukoplakia and normal tissue
[35]2019Diagnosis and PreventionCTAMedical/dental experience, psychosocial factors, demographics2401+Classification tree analysis (CTA) to identify population subgroups that are less likely to have an oral cancer examination (OCE)
[36]2004Diagnosis and PreventionANN, KLLC, PCASpectral data134+Classification and detection of invisible tissue alterations through autofluorescence spectroscopy applying PCA and ANN methods
[37]2003PrognosisLR, DTREETumor size, mode of invasion, and keratinization118+Three statistical methods for the prediction of lymph node metastasis in carcinoma of the tongue are compared
[38]2017Diagnosis and PreventionRFHistopathological images of OSCC 150+Automated technique for accomplishing the task of mitotic cell count from related histopathological images
[39] 2019Diagnosis and PreventionCNNHyperspectral images100+Proposed regression-based partitioned CNN learning algorithm for a complex medical image of oral cancer diagnosis
[40] 2019Diagnosis and PreventionCNNComputed tomography scan images441+Deep-learning image classification system for the diagnosis of lymph node metastasis on CT images
[41]2019Diagnosis and PreventionCNN, SVM, LDASpectral data1440+Classification method that discriminates tongue squamous cell carcinoma (TSCC) from non-tumorous tissue
[42]2018Diagnosis and PreventionSVMInfrared (IR) thermal imaging90+Automatic analysis by an entropy gradient support vector machine (EGSVM) using a digital infrared thermal imaging system
[43]2019PrognosisRF, DJU, LR, ANNPersonal details, tumor, and treatment characteristics 33,065+Describes a model using machine learning to help predict 5-year overall survival among patients with oral squamous cell carcinoma (OSCC)
[44]2015Potentially malignant oral lesions (pre-cancer)RVM, SVM, MLCImages of lip border150+Using robust macro-morphological descriptors of the vermillion border from non-standardized digital photographs with a probabilistic model (RVM) for solar cheilosis recognition
[45]2015Diagnosis and PreventionSVMSpectral data47+Classification of two oral lesions, namely oral leukoplakia (OLK) and oral squamous cell carcinoma (OSCC), was performed with SVM using different combinations of spectral features
[46]2019Diagnosis and PreventionLDA, SVMSpectral data34?Showed the specific IR spectral signature for OC salivary exosomes, which was accurately differentiated from HI exosomes based on detecting subtle changes in the conformations of proteins, lipids, and nucleic acids using optimized ANN
[47]2019Diagnosis and PreventionLRTissue microarray chips 105+The effectiveness of Aurora kinase A and Ninein interacting protein (AUNIP) in diagnosing OSCC was evaluated by machine learning
[48]2015PrognosisSVMProtein intensity30+Proteomeof whole saliva and salivary extracellular vesicles (EVs) from patients with OSCC and healthy individuals were analyzed. The proteomics data could classify OSCC with 90% accuracy
[49]2019Diagnosis and PreventionSVMPutrescine, glycyl-leucine, and phenylalanine31+With three-marker panel, consisting of putrescine, glycyl-leucine, and phenylalanine using a support vector machine (SVM) model that can discriminate paired cancerous (T) from adjacent noncancerous (AN) tissues
[50]2015PrognosisLRGene expression profiles486+The proposed network-driven integrative analytical approach can identify multiple genes significantly related to an OSCC stage
[51]2019PrognosisSVM, GB, LR, DTREEClinicopathologic data782+Machine learning improves prediction of pathologic nodal metastasis in patients with clinical T1-2N0 OCSCC compared to methods based on DOI
[52]2020PrognosisPLS-DA, OPLS-DASpectral data180+Spectral data on 180 tissues comprising tumor, margin, and bed from 43 OSCC patients were used to perform machine-learning models to identify malignancy status
[53]2018Therapy and quality of lifeDFGFamily, gene, compound, bile mutation, GWAS phenotype, OMIM phenotype, kidney mutation, and oral mutation400+Algorithm Medusa in parallel with binary classification was used in order to find potential compounds to inhibit oral cancer
[54]2013PrognosisANFIS, ANN, SVM, LRClinicopathologic and genomic data31+The results revealed that the prognosis is superior with the presence of both clinicopathologic and genomic markers
[55]2006Potentially malignant oral lesions (pre-cancer)WNNTEM images of collagen fibers from oral subepithelial region145+The trained network could classify normal fibers from less advanced and advanced stages of OSF successfully
[56]1996Diagnosis and PreventionANNIntra-oral smears348+A neural network differentiated between normal/non-dysplastic mucosa and dysplastic/malignant mucosa
[57]2011PrognosisLR, ANNMedical history211?Suggests the importance of routinely investigating PTI in OSCCs as useful marker of tumoral behavior and prognosis
[58]2003Diagnosis and PreventionPLS, ANNSpectral data97+The PLS-ANN classification algorithm based on autofluorescence spectroscopy at 330 nm excitation is useful for in vivo diagnosis of OSF, as well as oral pre-malignant and malignant lesions
[59]2020Potentially malignant oral lesions (pre-cancer)KNNDemographics, lesion characteristics, and cell phenotypes999+Cytopathology tools represent a practical solution for rapid PMOL assessment, with the potential to facilitate screening and longitudinal monitoring in primary, secondary, and tertiary clinical care settings
[60]2018Potentially malignant oral lesions (pre-cancer)CNNImages oral cavity170+CNN was implemented in the cloud and used for automatic image analysis and classification of pairs of images into “suspicious” and “non-suspect”
[61]2017Potentially malignant oral lesions (pre-cancer)SVMDIC images of oral exfoliative cells119+The selected morphological and textural features of epithelial cells are compared with the non-smoker (-ve control group) group and clinically diagnosed pre-cancer patients (+ve control group) using SVM classifier
[62]2020PrognosisDTREE, ANN, NB, KNNContrast-enhanced CT images40+A radiomic ML approach applied to PTLs is able to predict TG and NS in patients with OC and OP SCC
[63]2006Diagnosis and PreventionANN, PCASpectral data143+Spectral analyses were used to classify and discriminate among normal, pre-malignant, and malignant conditions on oral tissue. Sensitivity and specificity gave results of > 92% in PCA and ANN
[64]2017Potentially malignant oral lesions (pre-cancer)RF, SVM, KNNExfoliative cytology, histopathology, and clinical follow-up data364?Developed an exfoliative cytology-based method for quantitative prediction of cancer risk in patients with oral leukoplakia
[65]2005Diagnosis and PreventionSVM, RVMSpectral data325+The Bayesian framework of RVM formulation makes it possible to predict the posterior probability of class membership in discriminating early SCC from the normal squamous tissue sites of the oral cavity in contrast to dichotomous classification provided by the non-Bayesian SVM
[66]2012Potentially malignant oral lesions (pre-cancer)BC, SVMImages of normal and OSF oral mucosa119+Bayesian classification and support vector machines (SVM) to classify normal and OSF
[67]2020Diagnosis and PreventionSVM, DTREE, NCA, LDAH&E-stained microscopic images of squamous epithelial layer676+ML-based automatic OSCC classifier named as stratified squamous epithelial biopsy image classifier (SSE-BIC) to categorize H&E-stained microscopic images of squamous epithelial layer in four different classes: normal, well-differentiated, moderately differentiated, and poorly differentiated
[68]2015PrognosisDTREE, LR, ANNMedical history673-Determines the differences between the symptoms shown in past cases where patients died or survived oral cancer
[69]2018Diagnosis and PreventionSVMSpectral data186+Diffuse reflectance spectra were used to discriminate tumor from healthy tissue, and SVM models were used to classify them
[70]2015Diagnosis and PreventionLIR, DTREE, RF, TREEB, ANN, CCNN, PNN/GRNNMedical history1025-Data-mining model using probabilistic neural network and general regression neural network (PNN/GRNN) for early detection and prevention of oral malignancy
[71]2005Diagnosis and PreventionFNNAge, gender, smoking, alcohol, bcl-2, PCNA, surviving21?FNN were effectively used to analyze the relationship between oral leukoplakia and HPV infection
[72]2018PrognosisRF, DTREE, NB, LR, SVM, ANNPeptides and proteins40+Proteomics analysis of proteins in saliva in combination with machine-learning methods were applied to study prognosis classification
[73]2005PrognosisFNNAge, gender, smoking, alcohol, bcl-2, PCNA, surviving21?FNNs were used to build up a predictive model to study the relationship between HPV infection and different variables in the OSCC
“+” indicates low risk of bias, “-” indicates high risk of bias, and “?” indicates unclear risk of bias, CNN = Convolutional neural network, ANN = Artificial neural network, SVM = Support vector machine, RF = Random forest, DTREE = Decision tree, LR = Regression logistic, BC = Bayesian classifier, ANFIS = Adaptive neuro-fuzzy inference system, WNN = Wavelet neural network, CTREE = Conditional inference trees, BA = Bagging, LDA/QDA = Linear/quadratic discriminant analysis, RVM = Relevance vector machine, MLC = Mahalanobis classifier, LIR = Linear regression, TREEB = Tree boost, CCNN = Cascade correlation neural network, GB = Gradient boosting, NCA = Neighborhood component analysis, DJU = Decision jungle, CTA = Classification tree analysis, PLS-DA = Partial least square discriminant analysis, OPLS-DA = Orthogonal partial least square discriminant analysis, DFG = Data fusion graph, RFC = Randomizable filtered classifier, RT = Random tree, PCA = Principal component analysis, PLS = Partial least squares, NB = Naïve Bayes, KNN = K nearest neighbors, GP = Genetic programing, RT = Random tree, KLLC = Karhunen–Loeve linear classifier, XGBOOST = Extreme gradient boosting, PNN/GRNN = Probabilistic neural network /general regression neural network, FNN = Fuzzy neural network, KSTAR = Instance-based learner using an entropic distance measure, ROB = Risk of bias.
Table
Table 2. Statistical comparisons of reported performance metrics for support vector machine, neural network, and logistic regression algorithms. Metrics are shown according to the clinical contexts (diagnosis and prevention, prognosis, pre-cancer, and therapy and quality of life).
Table 2. Statistical comparisons of reported performance metrics for support vector machine, neural network, and logistic regression algorithms. Metrics are shown according to the clinical contexts (diagnosis and prevention, prognosis, pre-cancer, and therapy and quality of life).
Performance Metricas Mean (SD; n)
Accuracy
%		Sensitivity
%		Specificity
%		AUC
SVM85.83 (10.01; 19)86.45 (8.06; 17)88.20 (10.72; 15)0.83 (0.15; 9)
ANN75.72 (13.08; 8)76.90 (13.65; 9)84.59 (13.44; 8)0.69 (0.14; 5)
LR75.47 (12.67; 9)76.53 (13.68; 6)77.51 (10.78; 4)0.7 (0.14; 4)
ANOVAp-value = 0.037p-value = 0.058p-value = 0.270p-value = 0.213
Diagnosis and Prevention
SVM90.22 (5.79; 8)87.19 (4.69; 7)89.99 (7.80;7)0.95 (0.04; 3)
ANN84.03 (20.18; 2)84.51 (7.97; 5)83.28 (15.48; 5)0.6 (0.18; 2)
LR93.50 (9.19; 2)87.00 (0; 1)-------
ANOVAp-value = 0.570p-value = 0.761p-value = 0.343p-value = 0.039
Prognosis
SVM74.72 (13.19; 5)78.90 (11.95; 4)74.00 (26.87; 2)0.75 (0.16; 5)
ANN71.55 (11.69; 5)61.17 (7.80; 3)80.15 (4.44; 2)0.76 (0.09; 3)
LR68.21 (5.97; 6)73.05 (15.98; 4)74.35 (10.69; 3)0.7 (0.14; 4)
ANOVAp-value = 0.600p-value = 0.249p-value = 0.903p-value = 0.861
Potentially malignant oral lesions (pre-cancer)
SVM89.69 (2.65; 5)90.61 (5.38; 6)90.86 (3.34; 6)---
ANN---86.00 (0; 1)100 (0; 1)---
LR83.00 (0; 1)80.00 (0; 1)87.00 (0; 1)---
ANOVAp-value = 0.082p-value = 0.255p-value = 0.083---
Therapy and quality of life
SVM87.00 (0; 1)------0.89 (0; 1)
ANN80.00 (0; 1)---------
LR------------
ANOVA------------
---: Data not reported by authors of those articles.
Table
Table 3. Total number of studies according to the type of validation method implemented.
Table 3. Total number of studies according to the type of validation method implemented.
MethodCount
Hold out18
5-fold CV11
10-fold CV10
LOOCV7
3-fold CV3
7-fold CV2
4-fold CV2
9-fold CV1
Without validation (not mentioned)3
Total57
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Share and Cite

MDPI and ACS Style

López-Cortés, X.A.; Matamala, F.; Venegas, B.; Rivera, C. Machine-Learning Applications in Oral Cancer: A Systematic Review. Appl. Sci. 2022, 12, 5715. https://doi.org/10.3390/app12115715

AMA Style

López-Cortés XA, Matamala F, Venegas B, Rivera C. Machine-Learning Applications in Oral Cancer: A Systematic Review. Applied Sciences. 2022; 12(11):5715. https://doi.org/10.3390/app12115715

Chicago/Turabian Style

López-Cortés, Xaviera A., Felipe Matamala, Bernardo Venegas, and César Rivera. 2022. "Machine-Learning Applications in Oral Cancer: A Systematic Review" Applied Sciences 12, no. 11: 5715. https://doi.org/10.3390/app12115715

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop