MVGL-Net: A generalizable multi-view convolutional network for anterior segment OCT

Runzhi Zhou, Alice Z. Chuang, Robert M. Feldman, Luca Giancardo

Abstract: Highlights•We introduce a new multiple inputs model and pre-processing pipeline which contains fewer parameters and achieves comparable or better result than the top performing angle-closure diagnosis algorithm in a recent data challenge on an external independent dataset.•We perform a transfer learning strategy using our pre-trained model for a Plateau iris configuration diagnosis task, which to our knowledge, has not been attempted before with a deep learning image-based automated algorithm. The results show that the approach proposed has a better generalizability than comparable methods.•We compare and contrast different number of B-scans and models of increasing complexity finding that less complex models with multiple B-scan perform best.