Rare eye disease characterization using AI and multi-modal fusion

4 Mai 2023

---

Catégorie : Post-doctorant

---

Post-doctoral fellowship

Rare eye disease characterization using AI and multi-modal fusion

Clinical context

Retinitis Pigmentosa (RP) is an Inherited Retinal Dystrophy (IRD) and a leading cause of blindness in children and young adults. RP is typically characterized by night blindness, progressive reduction of the visual field and loss of central visual acuity. The prevalence is 1/4000 with approximately 16,000 patients in France and 90,000 in Europe. Major therapeutic break- throughs have occurred with gene replacement as well as mutation-specific cell-driven therapies and neuro-protective approaches. RP is emblematic of extremely high levels of heterogeneity. Very few reports address thoroughly the prediction challenge of causative genes using Artificial Intelligence (AI) techniques. The current under-representation of IRD in the AI literature reflects the well-known rare disease challenges: very high-level clinical heterogeneity, extreme biological complexity, small cohorts, scattered expertise centers and registries...

 

Figure 1: Multi-modal imaging for RP diagnosis. From left to right: color fundus photography, spectral domain optical coherence tomography, fundus auto-fluorescence.

RaReTiA project

Funded by the French National Research Agency, the RaReTiA project aims at creating a de novo French Rare Eye Diseases Database (FREDD) as a repository embedded in France Cohortes (FC) to address the scientific challenges of applying AI to RP research. Further objectives include multi-modal deep learning approaches to decipher RP clinical categories and sub-categories, improve RP early detection, grading and stratification as well as radiomic data mining techniques to predict phenotype-genotype associations at large, from genomic variant to biological systems.

Job description

The post-doctoral fellowship deals with investigating innovative AI methods for rare eye disease characterization. The aim will be to develop approaches able to improve diagnosis, classification and grading. A strong focus will concern the fusion of multi-modal data for the prediction of key elements in the management and follow-up of patients. In particular, the challenge will be to fuse 3 distinct imaging types (Fig.1) with contextual information and genetic variants.

Environment

- start date / duration: october 2023 for 12 months (renewable)

- laboratory: LaTIM1 UMR 1101, Inserm 1http://latim.univ-brest.fr