Medical Image Analysis

Overview

Medical image analysis is the computational backbone of my research. Before building longitudinal or predictive models, we need robust methods for understanding what is in a single image: detecting lesions, quantifying biomarkers, segmenting anatomical structures. In my work the primary modalities are ophthalmological.

Imaging modalities

Colour Fundus Photography

Wide-field retinal photographs capture the optic disc, macula, vessels, and peripheral retina. DR grading, vessel segmentation, and optic disc detection are well-established tasks. I use fundus images as the primary modality in several longitudinal and progression prediction pipelines.

Optical Coherence Tomography (OCT)

OCT provides cross-sectional depth profiles of retinal layers with micrometre resolution. Retinal layer segmentation, drusen volume measurement, and detection of fluid (IRF/SRF/PED) are key tasks for AMD monitoring. OCT is the workhorse modality in the MARIO challenge.

Fluorescein Angiography (FA)

FA captures vascular leakage and neovascularisation patterns that are invisible on fundus photography or OCT. Incorporating FA into multi-modal fusion models adds complementary disease activity information.

Core tasks

Related themes

• Multi-modal Learning
• Self-Supervised Learning
• Benchmark & Challenges