Deep-Learning on Rachid Youven Zeghlache

Explainable AI for Health

Thu, 01 Jan 2026 00:00:00 +0000

Overview

Deep learning models used in clinical settings must not only be accurate — they must be interpretable. Clinicians need to understand why a model predicts high-risk progression to trust and act on that prediction. Explainable AI (XAI) provides tools to open the black box and reveal which input features drive each decision.

Techniques I apply

Saliency maps

Gradient-based saliency methods (GradCAM, Integrated Gradients, SHAP for images) highlight which spatial regions of a retinal image or OCT scan most influenced the model’s prediction. A heatmap overlaid on a fundus image showing drusen, haemorrhages, or neovascularisation has direct clinical meaning.

MARIO AMD Progression Challenge

Sun, 01 Sep 2024 00:00:00 +0000

Overview

The MARIO AMD Progression Challenge was held at MICCAI 2024 and focused on the automated assessment of Age-related Macular Degeneration (AMD) progression using deep learning.

We are pleased to announce the publication of our comprehensive analysis of the challenge results, and the corresponding dataset is now publicly available for the research community.

Research Focus

This challenge addressed two key clinical questions:

Progression prediction: Will a patient’s AMD progress over the next 12 months?
Visual acuity change: Will the patient’s visual acuity improve, stabilise, or worsen?

Dataset

The MARIO dataset provides: