Table 2.
Work done by different researchers
| Ref No. | Disease Detected | Dataset Used | Technology/Approach | Key Outcomes |
|---|---|---|---|---|
| [6] | COVID 19 | COVID-19 records from 2,875 patients in three hospitals. 34 symptoms, with key ones as apnea, cough, fever, and CVD. |
Frequency-based feature selection with set thresholds. Apriori algorithm for symptom-outcome association. |
Symptom Associations: Recovery: fever, apnea, cough. Death: apnea, weakness, CVD, ventilator use. No causality, limited provider trust, lacks symptom progression tracking. |
| [12] | Cardiovascular Disease | Cleveland Heart Disease dataset (303 instances, 14 features) | Used SVM, KNN, AdaBoost, Gaussian Naive Bayes for heart disease prediction. Applied XAI for feature selection and model weight optimization. |
Achieved 82.5% accuracy with SVM. Enhanced interpretability for clinical decision-making. Small dataset and limited attributes reduce robustness. Reliance on a single dataset affects generalizability. |
| [13] | Parkinson’s Disease | 642 DaTSCAN SPECT images (430 PD, 212 non-PD | VGG16 CNN with transfer learning for classification. LIME for visual explanations of image influencing decisions. |
Accuracy: 95.2%, Sensitivity: 97.5%, Specificity: 90.9%. Aids early PD diagnosis and clinical decision-making. Class imbalance. Limited generalizability and dependence on image quality. |
| [14] | General Disease | General medical datasets including pneumonia, BSI, AKI, and ICU data | Explain ability approaches like LIME, SHAP, Grad-CAM,. Evaluation using AUROC and sensitivity analysis | Resource-intensive, with concerns over legal and ethical uncertainty. Risk of bias in data, impacting model decisions. |
| [15] | COVID-19 | COVID-19 dataset (50,000 + patients, from May to October 2020). | Used MLP and Random Forest for severity prediction (high, medium, low). Integrated LIME for improved model interpretability. |
80% accuracy with both MLP and RF. Real-time assessments available via mobile and web apps. Dataset limited to a specific region and time period. Performance variability in medium severity cases, potential overfitting |
| [18] | Review on Healthcare | 150 Articles in Healthcare and interpretability Models. | LIME, Transfer Learning | Systematic review of 53 articles, categorizing XAI methods like SHAP, LIME, and Grad-CAM. Discusses applications to diseases like brain tumors, COVID-19, and chronic kidney disease. |