AI for Ophthalmology
summary
A retrospective two-year cohort of dry AMD patients is used: baseline OCT images are fed to convolutional neural nets with a proprietary normalization step, trained after trying thousands of architectures on a Dell Precision workstation with three Nvidia GV100s. The model strongly predicts conversion to wet AMD (no AUC given) and localizes its signal to the choroid beneath the retina, suggesting a subtle texture biomarker humans miss; larger-scale validation is currently ongoing.
AI-written summary · about
chapters (7)
related talks
37mAI in medicine: where do we stand?
10mConnected eyes
44mAI in the Clinic: Prediction is Better Than Cure
28mAI in healthcare - Deep learning for precision psychiatry
49mBuilding AI That Learns and Adapts: A Case Study in MRI Diagnostics
36mAI in Healthcare
55mDeep Learning in Medical Image Diagnostics
17mAI Innovation in Health