bilha analytics

IRFundusSet - Large heterogeneous retinal fundus dataset

TLDR;

Table: Composition of IRFundusSet

Availability and quality of training data is a common challenge when developing AI models. Obtaining comprehensive and sufficiently sized datasets of health-related datasets is non-trivial.

A current challenge with retinal fundus datasets is the fragmentation of publicly available datasets. These datasets often differ significantly in their data organization, archiving methods, and in the definition of disease labels. In addition, the definition of a “healthy” or “normal” eye can vary considerably between datasets. Some may include images with minor, non-vision-threatening conditions as “normal,” while others may have stricter criteria. This inconsistency can restrict the overall diversity of data available for training AI models and often necessitates substantial additional curation efforts by researchers.

The Integrated Retinal Fundus Set (IRFundusSet) aims to consolidate, harmonize, and curate several existing public retinal fundus image datasets into a more unified and accessible resource.

Accessing IRFundusSet