Related Experiment Video
Updated: Feb 17, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.6K
Unveiling the black box: Explainable transfer learning for ocular disorder diagnosis
Zaib Un Nisa1, Arfan Jaffar1, Sohail Masood Bhatti1
1Department of Computer Science and Information Technology, The Superior University, Lahore, Pakistan.
Digital Health
|February 16, 2026
Summary
Transfer learning models show promise for diagnosing eye diseases like diabetic retinopathy. However, explainable AI is crucial for reliable model selection, especially for conditions like glaucoma.
Area of Science:
- Ophthalmology
- Artificial Intelligence
- Medical Imaging
Background:
- Multiclass ocular disease diagnosis is critical for patient outcomes.
- Transfer learning (TL) and deep learning offer potential for automated diagnosis.
- Evaluating model reliability and interpretability is essential for clinical adoption.
Purpose of the Study:
- To systematically evaluate transfer learning models for multiclass ocular disease diagnosis.
- To assess the reliability of these models using explainable artificial intelligence (AI).
Main Methods:
- Eight pretrained convolutional neural network (CNN) models were evaluated on a public dataset.
- Performance metrics included accuracy, precision, recall, and F1-score.
- Explainability techniques (Grad-CAM, LIME, SHAP) were employed.
Main Results:
- Models achieved high accuracy for diabetic retinopathy and cataract detection (DenseNet121, XceptionNet).
- Glaucoma diagnosis showed weaker results, suggesting a need for segmentation-based approaches.
- Explainability revealed significant differences in model attention despite similar accuracy.
Conclusions:
- Accuracy alone is insufficient for trustworthy medical AI.
- Explainable AI is vital for selecting reliable diagnostic models.
- EfficientNetB3 demonstrated a good balance of performance and interpretability; glaucoma diagnosis needs advanced pipelines.
