Corneal-disease detection
Leading a healthcare vision project from clinical data collection to weakly supervised detection.
Role: Tech Lead, ARAS Labs, 2021 to 2022. I led three engineers.
The problem
The project aimed to detect corneal diseases, including keratoconus, from clinical data. This meant working with real hospital data and with several experts whose labels could disagree.
What I did
- I led the technical and research work, directing the engineers on implementing the algorithms and on the pipeline from data processing to predictions.
- I coordinated data collection with ophthalmic surgeons and annotation by ophthalmologists.
- I designed the label aggregation that turns labels from several experts into a single ground-truth label.
- I proposed a weakly supervised learning approach to detection.
- I worked with the hospital to anonymise clinical data and set up an on-site processing environment, so identifiable data stayed on hospital premises.