Abstract

Diabetic retinopathy (DR) is the leading cause of vision loss in working-age adults. Optical Coherence Tomography-Angiography (OCTA), a non-invasive functional extension of optical coherence tomography, is emerging as the potentially most promising technique for diagnosing DR [1, 2]. Recent technological advances allow for ever-increasing large OCTA scans of the retina including its periphery [3, 4]. Early detection of DR has long been implemented using artificial intelligence (AI)-algorithms, ranging from fundus photography to OCTA volume data. Supervised learning-based AI algorithms, which learn based on annotated data, require labor-intensive labeling by expert graders. Multiple instance learning (MIL) is a supervised learning technique that allows for binary classification in noisy datasets. With our work here, we circumvent the necessity for precise labeling through our novel proposed multiple instance learning (MIL) network approach, MIL- ResNet14.

© 2023 SPIE