Explainable Deep Learning for Disease Diagnosis and Medical Imaging

Project Description

Deep learning now matches or exceeds clinicians on many diagnostic and screening tasks, yet adoption stalls when models cannot explain their decisions or generalize beyond curated benchmarks. Across a range of conditions—from sight-threatening eye disease to chronic kidney and bone disorders—InfoLab at Sungkyunkwan University (SKKU) builds diagnostic systems that are accurate, robust, and interpretable, designed to earn clinical trust rather than just top a leaderboard.

This project unifies our work on explainable AI (XAI) for disease diagnosis and medical imaging beyond neurodegenerative disease, combining transfer learning, hybrid CNN–Transformer architectures, and ensemble methods with explanations clinicians can inspect.

Core Research Themes and Contributions

1. Diabetic Retinopathy Grading from Retinal Images

We develop systems that detect multiple lesions and grade diabetic retinopathy severity from fundus images. Our hybrid models leverage transfer learning and CNN–Transformer synergy within a robust deep-ensemble architecture, improving reliability across imaging conditions and supporting earlier, automated screening for a leading cause of preventable blindness.

2. Explainable Chronic Kidney Disease Detection

We design hybrid explainable AI pipelines for chronic kidney disease (CKD) detection and prediction, pairing ensemble learning with case-based reasoning so that each prediction is accompanied by interpretable, clinically meaningful evidence.

3. Computer-Aided Osteoporosis and Skin-Cancer Assessment

We contribute datasets and models for osteoporosis screening from bone densitometry, and study staging of melanocytic skin neoplasms using high-level, pixel-based features—extending explainable diagnosis to bone health and dermatology.

4. Explanation as a First-Class Output

Across all tasks, predictions are paired with interpretability tools (e.g., saliency/attribution and case-based explanations) that surface why a model reached its conclusion, enabling clinicians to validate, override, or trust the system.

Project Objectives

• Build accurate, robust diagnostic models across imaging and clinical-data modalities for multiple diseases.
• Make every prediction explainable and clinically verifiable, not a black box.
• Exploit transfer learning, hybrid CNN–Transformer designs, and ensembles for reliability under real-world variation.
• Release datasets, benchmarks, and interpretable modules to support reproducibility and clinical translation.

Research Impact

By extending explainable deep learning to retinopathy, kidney disease, osteoporosis, and skin cancer, InfoLab (SKKU) broadens trustworthy medical AI well beyond a single disease. The result is a portfolio of screening and diagnosis tools that prioritize transparency and robustness—qualities essential for AI that clinicians can adopt in everyday practice.