Table 1 Characteristics of XAI-related studies in disease prediction (*The journal name may be abbreviated or formatted based on its indexing style. **The country information may refer to either the authors' affiliated country or the country where data was collected for the disease prediction model.)
Paper | Journal* | Country** | Disease | Data modality | Data type | Data Availability | AI Model | XAI method | Contribution |
|---|---|---|---|---|---|---|---|---|---|
[12] | Healthcare Technology Letters | Bangladesh | Diabetes | Physiological and medical measurements | Numerical | Yes | DT SVM RF LR KNN | LIME SHAP | ML-enhanced platform for diabetes diagnosis accessible via web and mobile app |
[13] | Sensors journal | Bangladesh Republic of Korea | Ischemic Stroke | Electroencephalography (EEG) | EEG Signal | No | XGBoost LGBM Adaptive Gradient Boosting | Eli5 LIME | EEG-based ML model for stroke prediction, focusing on brain wave analysis |
[14] | Computational Intelligence and Neuroscience | Bangladesh Saudi Arabia | Acute Lymphocytic Leukemia (ALL) | White Blood Cells image | Image | Yes | Pretrained CNNs using Transfer Learning | LIME | InceptionV3 and LIME-based XAI model for automated Acute Lymphoblastic Leukemia detection |
[15] | BMC Med Inform Decis Mak | Brazil | Sarcoidosis | Forced Oscillation Technique (FOT) exams | Numerical | Yes | KNN SVM AdaBoost RF LGBM XGBoost DT LR | Genetic programming (for intelligible expressions), Feature Importance Evaluation | Proposes a genetic programming approach with XAI for diagnosing sarcoidosis from respiratory function tests |
[16] | JACC: CARDIOVASCULAR IMAGING | USA Canada Israel Switzerland | Obstructive Coronary Artery Disease (CAD) | Single-Photon Emission Computed Tomography (SPECT) Myocardial Perfusion Imaging (MPI) | Numerical | No | CAD-DL | XAI | Explainable DL model for rapid and accurate Coronary Artery Disease detection from cardiac images |
[6] | Computers in Biology and Medicine | India | Parkinson’s Disease (PD) | Gene Expression Data | Numerical | Yes | LR SVM DT RF KNN NB | SHAP | Applies ML and XAI to early Parkinson’s Disease detection, highlighting significant biomarkers |
[17] | Diagnostics | Saudi Arabia Egypt | Colon Cancer | WCE colon image dataset | image | Yes | Heterogenic stacking DL integrated with pretrained CNN models [VGG16, InceptionV3, Resnet50, DenseNet121) | XAI | Introduces a heterogenic stacking DL model with XAI for improved colon cancer prediction |
[18] | Journal of Medical Systems | China | Hepatitis | Mixture of integer and real value attributes | Numerical | Yes | LR DT KNN XGBoost SVM RF | SHAP LIME PDP | Presents an XAI framework aimed at enhancing hepatitis diagnosis and informing clinical decisions |
[19] | Computer Methods and Programs in Biomedicine | South Korea USA Egypt | Parkinson’s Disease (PD) | SC, BS, MH, M, and NM | Time series data | Yes | SVM RF ETC LGBM SGD | LIME SHAPASH | Proposes a multi-modal time series ML pipeline with explainability for Parkinson’s Disease progression prediction |
[20] | IEEE Transactions on Biomedical Engineering | Algeria France | Breast Cancer | Clinicopathological data | Categorical and continuous | Yes | CatBoost | LIME | Outlines an explainable ML prognosis model for cancer metastasis, aiding personalized treatment decisions |
[21] | Nature portfolio Scientific reports | Spain Egypt Republic of Korea | Alzheimer’s Disease | Multimodal (PET, CSF, Cognitive Scores, Genetics, Lab Tests, Medical History, MRI, Neurological Exam and others) | Both | Yes | RF DT Binary Classification | SHAP Fuzzy | Developes a system that diagnoses and detects the progression of Alzheimer’s Disease by taking into consideration different modalities |
[22] | Nature research Scientific Reports | Canada | COVID-19 | Chest X-ray | Image | Yes | CNN-Based Model: Proposed COVID-Net | GSInquire | Introduces COVID-Net which is a deep convolutional neural network that detects COVID-19 cases using explainability methods |
[5] | Journal of Biomedical Informatics | Slovakia Japan | Colorectal Cancer | Histopathological data | Image | Yes | CNN | Explainable Cumulative Fuzzy Class Membership Criterion (X-CFCMC) | The classifier uses histopathological data and can predict 8 varieties of colorectal cancer |
[23] | Annals of Translational Medicine | China | Fenestral Otosclerosis | Temporal bone high-resolution computed tomography (HRCT) slices | Image | Yes | CNN-Based Model: Proposed Otosclerosis Logical Neural Network Model | Visualization of learned deep representations | Enhances the diagnosis of fenestral OS using temporal bone high-resolution computed tomography (HRCT) slices and DL |
[24] | Computers in Biology and Medicine | India | Parkinson’s Disease | Single-photon Emission Computed Tomography (SPECT) DaTSCANs | Image | Yes | VGG16 | LIME | Develops an improved accuracy DL model for early diagnosis and provided visual markings generated by the model to aid medical practitioners |
[25] | The Journal of Supercomputing | India | Heart Diseases | Age, Resting Blood Pressure, Exercise Induced Angina, Fasting Blood Sugar, Maximum Heart Rate, Serum Cholesterol and 7 other features | Numerical | Yes | XGBoost | SHAP LIME PDP DALEX | Works on the reduction of dimensionality using XAI while maintaining the model’s accuracy |
[26] | Genes | India Egypt Morocco Qatar | Cervical Cancer | Vaginal Swab Samples [Microbial Data (16S rRNA Sequencing)] | Genomic Sequence | Yes | RF | SHAP | Uses specific microbial patterns commonly found in cervical cancer to create personalized medicine |
[27] | Cancers | Pakistan Saudi Arabia United Arab Emirates United Kingdom Saudi Arabia | Lung Pulmonary Disease | Chest Radiographs | Image | Yes | CNN-based transfer learning with RestNet50 | LIME | Shows improved accuracies and explanations in interpreting pulmonary diseases using chest radiographs |
[28] | Radiation Oncology | Germany | Prostate Tumor | Multi-parametric MRI | Image | Yes | U-Net architecture CNN | Grad-CAM | Proposes an XAI framework that can identify tumor prostate tissues through images |
[29] | Computers in Biology and Medicine | Nepal and Australia | COVID-19, Pneumonia, and Tuberculosis | Chest X-ray | Image | Yes | CNN | SHAP LIME Grad-CAM | Uses a CNN model to detect lung diseases using CXR images with a focus on interpretability for clinicians |
[30] | Nature Communications | USA China | Kidney-related disease | Electronic health records (EHRs), medical imaging data, laboratory tests | Myocardial perfusion, wall motion, and wall thickening polar maps | No | GBT | SHAP | Creates and evaluates a transportable, XAI Model for Acute Kidney Injury Prediction |
[31] | Frontiers in Medicine | Italy France | Breast cancer | Clinical and cytohistological outcomes from patients’ medical records | Clinical outcomes, therapy-related information | Yes, by request | SVM RF NB XGBoost | SHAP | XAI approach reveals critical factors affecting breast cancer IDEs at 5 and 10-year post-diagnosis, aiding personalized patient care |
[32] | Frontiers in Cardiovascular Medicine | Finland | Cardiovascular disease | Medical records | Numerical and nominal features | Yes | Ensemble Tree algorithm | SHAP | Creates an XAI for Heart Failure Prognosis, Balancing Clinical Insight with Predictive Precision |
[33] | Frontiers in Neuroscience | Italy USA | Neuroscience | Imaging, electrophysiological recordings, clinical assessments | EEG, spike data | No | ML algorithms | LIME | Advances AI and XAI for Enhanced Brain Function and Neurostimulation Insights, Highlighting Transparent Models and Data Competitions |
[34] | NPJ Digit Medicine | UK | Mental health | Clinical information, patient records, and diagnostic data | patient characteristics, diagnostic indicators, and potentially text-based information | Yes | Deep neural networks, prediction and classification in psychiatric applications | SHAP | Presents the TIFU Framework for Clear, Interpretable AI in Mental Health, with a Focus on Clinically Aligned AI Comprehensibility |
[35] | Compute Methods Programs Biomed | Spain | Prostate cancer tissue | Specifically RNAseq data obtained from transrectal biopsies | gene expression data | Yes | KNN rpart (CART) RF | SHAP | Offers a ML Classifier Using Gene Expression Data and XAI for Precise Prostate Cancer Risk Prediction |
[36] | Clinical Medicine Insights Cardiology | UK and Sweden | Myocardial Infarction (MI) | Medical history, and demographic data | Physical and functional measures, and collection of blood, urine, and saliva | Yes, by request | LR XGBoost | SHAP | XGBoost with SHAP Values: A Promising Method for Predicting Myocardial Infarction Risk Across a Broad Population |
[37] | Nature Communications | Denmark | Acute Critical Illness | Electronic Health Records | Secondary healthcare data from four Danish municipalities | Yes, by request | TCN | LRP | Presents an xAI-EWS System for Predicting Acute Illnesses with EHRs, Offering Clear, Real-time Insights for Clinician Decision-making |
[38] | Nature Portfolio Scientific Reports I | Italy | Thyroid | Histological samples | Raman spectra | Yes, by request | ML algorithms RF XGBoost | SHAP | Combines Raman Spectroscopy and ML for Non-Invasive Thyroid Cancer Diagnosis to Potentially Lower Unneeded Surgeries |
[39] | European Federation for Medical Informatics | Greece | Preterm birth | Demographics, social and medical history, and obstetrics variables | Numerical, ordinal, and nominal features | No | LR SVM RF XGBoost | SHAP | Forecasts Preterm birth chances using demographic and medical data, providing predictions and insights for enhanced pregnancy screening |