Integrating Machine Learning with Medical Imaging for Human Disease Diagnosis: A Survey ^†

Abstract

Machine learning is revolutionizing healthcare by enhancing diagnosis and treatment personalization. This study explores ML applications in medical imaging, analyzing data from X-rays, CT, MRI, and ultrasound for early disease detection. It reviews key ML models, including SVM, ANN, RF, CNN, and other methods, demonstrating their effectiveness in detecting cancers such as lung and prostate cancer and other diseases. Despite their accuracy, these methods face challenges such as a reliance on large datasets and significant computational requirements. This study highlights the need for further research to integrate ML into clinical practice, addressing its limitations and unlocking new opportunities for improved patient care.

1. Introduction

Healthcare is rapidly evolving, with advancements in digital technologies such as 3D printing, robotics, and nanotechnology. These innovations offer significant opportunities to reduce human errors, enhance clinical outcomes, and ensure efficient long-term data tracking. From machine learning to deep learning, AI plays a crucial role in various medical domains, including disease diagnosis and prognosis, drug design and discovery, remote health monitoring, and robot-assisted surgery. With sophisticated algorithms and advanced analytical capabilities, AI, from machine learning to deep learning, is transforming traditional clinical practices, enabling smarter, more patient-centered approaches and facilitating the management of complex diseases [1].

Medical imaging has become an essential component in modern healthcare, playing a crucial role in assisting with medical diagnosis. With the continuous growth of medical image databases, it has become increasingly difficult to derive simple methods for object representation due to the complexity and variability of medical imaging data. Computer-based analysis of these images not only aids physicians in disease diagnosis but also serves as a powerful tool for therapeutic planning. In this context, machine learning techniques have emerged as effective solutions, enabling the development of intelligent systems that support doctors in diagnosing diseases, predicting risks, and ultimately saving human lives. These methods are now fundamental in the interpretation and processing of medical data, making them a cornerstone of decision support in clinical practice [2].

The integration of AI and machine learning in the medical domain relies on the exploitation of data from multiple modalities, including electronic medical records, medical imaging, genomics, and behavioral or environmental data. Among these sources, medical imaging plays a central place by providing essential diagnostic information for the detection and management of pathologies. With technologies such as X-rays, computed tomography (CT), magnetic resonance imaging (MRI), and ultrasound, healthcare professionals gain detailed visualizations of the human body, thereby facilitating diagnosis and treatment monitoring.

Machine learning techniques are widely used in the medical imaging research field as successful classifiers and segmentation algorithms. Medical image processing refers to a set of procedures aimed at obtaining clinically meaningful information from various imaging modalities, commonly for diagnosis or prognosis. In this process, the first step often involves segmentation, where structures of interest, such as tumors or organs, are isolated from the rest of the image. Then, feature extraction is performed to analyze these specific regions and extract essential information, which is used for classification and patient diagnosis. By combining segmentation and classification, these machine learning techniques enable more accurate results and facilitate medical diagnosis.

To gain a comprehensive understanding of how machine learning contributes to the diagnosis and prediction of various diseases, such as prostate cancer, lung cancer, liver disorders, heart conditions, and neurological diseases, it is crucial to examine the implementation and effectiveness of different techniques, like support vector machines (SVMs), artificial neural networks (ANNs), random forests (RFs), and convolutional neural networks (CNN), among others. In this study, we present a thorough review of the medical imaging datasets employed, detailing their feature extraction methods and classification strategies used for predictive analysis. We also compare the performance of these approaches based on evaluation metrics such as accuracy, sensitivity, specificity, area under the curve (AUC), precision, and recall.

This work provides an in-depth exploration of the impact of ML on healthcare through five main axes: the practical applications of ML in medical management, the multimodal data used in healthcare, the importance of medical imaging in modern diagnostics, applications of machine learning in medical imaging, and a comparative analysis of machine learning approaches applied to specific pathologies. These axes help identify current trends, highlight existing challenges, and open up perspectives for better integration of ML into healthcare systems.

2. Machine Learning Applications in Healthcare

Machine learning has the potential to revolutionize the healthcare industry by enabling more accurate diagnoses, personalized treatment plans, and improved patient outcomes. In this section, we will explore some of the ways machine learning is being used in healthcare today. Currently, machine learning is being used in a number of applications in healthcare, including those detailed below.

Treatment Design: Machine learning (ML) is crucial for personalizing treatment plans based on disease severity. By analyzing genetic information, medical history, and lifestyle variables, ML helps suggest optimal therapies. For example, in one study, KNN was used to recommend treatment plans for breast cancer patients, with options such as surgery, chemotherapy, hormonal therapy, biological therapy, and radiation therapy. KNN helps determine the most appropriate treatment by identifying the most similar “neighbors” (patients) in the dataset, enabling personalized recommendations based on similar cases [3].

Disease Diagnosis and Prognosis: Machine learning has shown remarkable potential in the field of medical diagnostics, particularly in detecting diseases such as skin cancer. By leveraging advanced algorithms, machine learning models can analyze medical images with high precision, often surpassing the diagnostic capabilities of experienced doctors. These models are trained on large datasets containing annotated images of skin lesions, allowing them to identify patterns and subtle characteristics that might be overlooked by human practitioners. The use of machine learning in detecting skin cancer, such as melanoma, enables faster and more accurate diagnoses, helping in early detection, which is crucial for successful treatment. Additionally, the continuous improvement of these algorithms, fueled by advancements in deep learning and neural networks, enhances their ability to handle complex image variations, making them highly reliable in clinical settings [4].

Drug Design and Discovery: Traditional drug development is hindered by high costs, long timelines, and inefficiencies. The complexity of big data from genomics, proteomics, and clinical trials adds further challenges. To address these issues, AI and machine learning are now used in various stages of drug discovery, including virtual screening, drug repositioning, and activity prediction. These techniques improve efficiency, accuracy, and speed in identifying therapeutic targets and developing new treatments [5].

Remote Health Monitoring: Numerous wearables and fitness gadgets can monitor individual health. These gadgets use machine learning to monitor health, analyze data, and provide users with insights to determine their health status [6]. For example, an intelligent system can predict and manage diabetes by analyzing vital signs like blood glucose and blood pressure in real time, alerting healthcare professionals when needed [7].

Robot-Assisted Surgery: Robotic-assisted surgery (RAS) offers several advantages, including precision, minimal invasiveness, and faster recovery, but it is costly. Given the high expense of robotic systems, it is essential to maximize their use. One way to achieve this is by efficiently managing the duration of each robot’s involvement in a procedure. This study explores the use of machine learning models, such as boosted regression tree, to predict surgery durations more accurately. By improving these predictions, hospitals can optimize robotic surgery scheduling, thereby enhancing resource utilization [8].

In this work, we focus on Disease Diagnosis and Prognosis, which represents one of the most promising applications of machine learning. With powerful algorithms, machine learning is capable of detecting diseases such as prostate and lung cancer with an accuracy often surpassing that of an experienced physician. These automated systems integrate complex and diverse data to provide reliable and rapid diagnoses, thereby reducing the risk of human error.

Machine learning systems analyze medical data, including images (X-rays, MRIs, CT scans), laboratory results, and electronic medical records. These tools assist doctors by quickly identifying abnormalities such as tumors, fractures, or other often subtle issues. For example, machine learning can detect anomalies in medical images with a level of accuracy that helps radiologists confirm or refine their conclusions.

3. Data Modalities in Healthcare

In healthcare, data comes from multiple modalities, such as electronic health records (EHRs), medical imaging, wearable devices, genomic data, sensors, environmental data, and behavioral data. Each of these sources offers a unique perspective on improving healthcare, as illustrated in Figure 1.

Among these modalities, medical imaging plays a vital role in smart healthcare. It provides essential diagnostic information and supports the management of various pathologies. With advanced imaging technologies, it captures detailed images of the human body, helping professionals visualize anatomical structures, detect abnormalities, and monitor treatment progression. Integrated with digital technologies and data analysis, medical imaging enhances the efficiency, accuracy, and accessibility of healthcare services, making it a cornerstone of modern medical practice [9].

4. Medical Imaging Types

Medical imaging techniques like X-rays, CT, MRI, and ultrasound provide vital data, visualizing tissues and tumors to aid in diagnosing and treating various conditions. These methods assess organs such as the brain, bones, and soft tissues, offering a clear view of internal anatomy essential for patient care [10].

• X-ray: Discovered in 1895, X-rays produce 2D images based on the variable absorption of radiation. They are used to detect fractures, lung diseases, cancers, osteoporosis, and dental cavities. Fast and cost-effective, they require precautions due to radiation exposure [10].
• CT (computed tomography): Developed in 1971, CT enhances X-rays by producing 3D images with better resolution and contrast. It is widely used in cardiology, oncology, and vascular radiology to detect tumors, lesions, and evaluate cancer spread, particularly breast cancer. Unlike X-rays, CT differentiates [10].
• MRI (magnetic resonance imaging): MRI, invented in the 1980s, visualizes abnormalities undetectable by other techniques like CT scans or X-rays. Unlike these methods, which map tissue density, MRI maps proton energy propagation. It uses a powerful magnetic field to align protons and a radiofrequency current to disrupt their balance, capturing images of soft tissues rich in water and fat [10].
• Ultrasound: Medical ultrasound uses a probe emitting ultrasonic signals to visualize internal body tissues such as muscles, tendons, and ovaries in real time. It allows for the observation of tissue structure, blood flow, and anomalies like cysts without harmful radiation. Its advantages include safety, portability, and real-time imaging, but it requires a qualified professional and provides a limited field of view [10].

5. Applications of Machine Learning in Medical Imaging

The main applications of ML in the domain of medical imaging include classification, detection, and segmentation. These applications all contribute to improving the accuracy and efficiency of medical diagnosis by providing clinicians with advanced tools to interpret medical images.

• Segmentation: In medical image analysis, segmentation is essential for enhancing diagnostic accuracy. It isolates relevant regions from surrounding tissues, allowing clinicians to focus on key areas for better assessment. This is particularly important for detecting and monitoring diseases like prostate and lung cancer, as it highlights anatomical and pathological structures with clarity [2].
• Classification: Classification is vital for accurate diagnosis, as it categorizes segmented regions such as distinguishing between benign and malignant tumors. Using machine learning, it analyzes features from images to assess disease severity and support treatment decisions. Alongside segmentation, classification ensures reliable and precise medical diagnosis [2].

6. ML Methods and Models: Previous Work

The application of machine learning techniques for disease diagnosis and prediction using medical imaging has been extensively studied, emphasizing their important role in improving healthcare outcomes. Many studies have explored the various methods to enhance diagnostic accuracy and predictive performance across different conditions.

Table 1. Relevant studies on image-based disease classification.

Type of Disease	Publisher	Publication Year	Dataset	Methodology	Performance Metrics
Lung Cancer [11]	Journal of Digital Imaging	2016	LIDC-IDRI (CT Scan)	Shape-based, margin-based, and texture-based feature extraction techniques; feature selection using A-z and p-values; SVM classifier.	AUC (Az) performance: Configuration 1: 95% Configuration 2: 88% Configuration 3: 84%
Lung Cancer [12]	Heliyon	2024	LIDC-IDRI (CT Scan)	Improved random walker with artificial neural network and random forest classifier.	Accuracy: Random Forest classifier: 99.6% ANN: 94.8%
Prostate Cancer [13]	European Radiology	2019	ProstateX challenge (MRI)	Multi-channel 3D convolutional neural network, cross-validation	AUC: 89% Sensitivity: 88.6% Specificity: 90.5%
Prostate Cancer [14]	Romanian Journal of Morphology and Embryology	2020	Included 439 images from 83 prostate cancer patients	GoogleNet AlexNet	Accuracy: GoogleNet: 60.9% AlexNet: 61.17%
Prostate Cancer [15]	Sage Journals	2019	ProstateX challenge	VGG-16 InceptionV3	AUC: VGG-16: 83% InceptionV3: 81%
Liver Disease [16]	Scientific African	2023	Kaggle Liver CT Scan Dataset and real-time images collected from Government General Hospital Vijayawada	Region-growing, Dens-Net CNN	accuracy: 98.34% sensitivity: 99.72% recall: 97.84%
Liver Disease [17]	Materials Today: Proceedings	2023	MVISS 3.0 T and gaggle liver patient dataset	LBP, GLCM, SVM	-
Heart Disease [18]	Computers, Materials & Continua	2024	Sunnybrook Cardiac Data (HNET-DSI) Kaggle datasets of cardiac disease patients (HNET-DSII) Github-datasets of cardiac disease patients (HNET-DSIII)	Heart-Net	Accuracy of: HNET-DSI: 92.56% HNET-DSII: 93.45% HNET-DSIII: 91.89%
Brain Disease [19]	Neural Computing and Applications	2020	Brain MRIs	BN-AlexNet-ELM-CBA	Sensitivity: 97.14% Specificity: 95.71% Accuracy: 96.43%

summarizes selected studies, including disease type, publisher, year, dataset, methodology, and performance metrics, allowing for a direct comparison of ML-based detection and classification approaches.

6.1. Lung Cancer

In this subsection, articles related to lung cancer diagnosis using ML methods are discussed. Dhara et al. developed a radiomic feature extraction framework focusing on texture and shape descriptors using feature selection based on A-z and p-values. A support vector machine (SVM) classifier was applied to distinguish benign from malignant lung lesions on DICOM images. The dataset was categorized into five malignancy ranks (1 to 5), and three configurations were tested to evaluate classification performance [11]. Similarly, Nair et al. proposed a method combining enhanced random walker segmentation with artificial neural networks (ANNs) and random forest classifiers, boosting detection accuracy [12].

6.2. Prostate Cancer

Several machine learning and deep learning approaches have been explored for prostate cancer diagnosis using multiparametric MRI (mpMRI) data. Aldoj et al. proposed a semi-automatic classification method using a multi-channel 3D convolutional neural network (3D CNN) to distinguish cancerous from non-cancerous lesions. Their model combined different mpMRI sequences, achieving the best performance with ADC, DWI, and K-trans parameters [13]. Marinescu et al. employed pre-trained deep networks, such as AlexNet and GoogleNet, to automatically classify prostate cancer tissue slices stained with Hematoxylin and Eosin (H&E) into four Gleason patterns [14]. Similarly, Chen et al. developed a deep convolutional neural network (DCNN) enhanced by transfer learning and data augmentation, such as rotation, to improve the classification of malignant lesions on mpMRI images [15].

6.3. Liver Disease

Several studies have explored liver disease diagnosis using medical imaging and machine learning. Prakash et al. proposed a model using a DenseNet CNN trained on the Kaggle Liver CT Scan Dataset, enhanced by region-growing segmentation, with validation on real-time data from Government General Hospital Vijayawada, achieving promising results [16]. In a similar study, Nanda Prakash et al. developed a system for liver cirrhosis prediction using MRI images. They isolated affected areas using ROI discovery, followed by feature extraction with the LBP and GLCM techniques, and trained an SVM classifier to differentiate healthy and cirrhotic cases, providing an efficient solution [17].

6.4. Heart Disease

Alsekait et al. propose Heart-Net, a multi-modal deep learning framework to improve the diagnosis of cardiovascular diseases. This framework integrates cardiac MRI and ECG signals, employing a 3D U-Net architecture for MRI analysis and a Temporal Convolutional Graph Neural Network (TCGN) for feature extraction from ECG data. The extracted features are combined using an attention mechanism to focus on the most relevant information for diagnosis. The classification task is carried out using an optimized version of the TCGN, showing enhanced diagnostic performance [18].

6.5. Brain Disease

Lu et al. conducted a study aimed at classifying brain MRI images into two categories: normal and abnormal. In the first step, they utilized a modified version of the AlexNet convolutional neural network (CNN) to automatically extract deep features from the MRI images. In the second step, these features were classified using an Extreme Learning Machine (ELM) to distinguish between normal and abnormal cases. To enhance classification accuracy and avoid poor convergence due to randomly initialized weights, the study introduced an optimization of the ELM using the Chaotic Bat Algorithm (CBA). This metaheuristic, inspired by the echolocation behavior of bats, is augmented with chaotic sequences for better exploration and convergence [19].

The integration of machine learning (ML) into disease detection has demonstrated remarkable progress, enhancing diagnostic accuracy and efficiency. However, several key considerations must be addressed to ensure the successful translation of ML models into clinical practice. Despite its success, ML adoption in medical imaging still faces several challenges:

• Efficient Feature Selection Technique: A computationally efficient feature selection method is needed to eliminate data cleaning steps while enhancing the accuracy of disease prediction [11].
• Small Sample Sizes: The main challenge faced in most studies lies in insufficient data to train the model. To train advanced models such as ANN or random forest, a sufficiently large sample is required to avoid overfitting and improve generalization [12].
• Resource-Intensive Methods: Training and deploying deep learning models demands significant computational resources, often requiring specialized hardware such as GPUs or TPUs. Research like [13,16,18,19] highlights the substantial processing power needed, which can hinder widespread implementation.
• Dependence on Imaging Quality: The performance of the model is highly influenced by the quality of input images. Research like [14,15,18] emphasizes the importance of high-quality, standardized imaging to ensure accurate and reliable predictions.

7. Conclusions and Perspectives

In conclusion, regarding disease diagnosis in general, accuracy is critical for planning effective treatment and ensuring the well-being of patients. Aiming at illuminating how machine learning techniques combined with advanced imaging technologies work in various disease diagnosis areas, the current study has been divided into several sections; the first section is Machine Learning Applications in Healthcare, followed by Data Modalities in Healthcare, Medical Imaging Types, and, finally, Applications of Machine Learning in Medical Imaging. Previous works on multiple diseases and a comparative analysis of different techniques with the datasets used, as well as the results of the machine learning method applied in terms of multiple parameters such as accuracy, sensitivity, specificity, and area under the curve, are discussed. According to the findings of this study, ML approaches combined with medical imaging technologies are essential for disease detection and improving diagnostic accuracy. However, despite ML’s promising capabilities, several challenges persist, including small sample sizes that lead to overfitting, resource-intensive methods, and the dependency on high-quality standardized imaging data. Moreover, the computational demands for training and deploying ML models remain high, which can limit their widespread adoption. Future research must address these issues by increasing dataset diversity and optimizing models to reduce resource dependencies. These efforts will be essential in ensuring a broader clinical implementation of machine learning systems, ultimately improving patient outcomes and optimizing healthcare.