Ai-Based Prediction Of Locs Iii Cataract Grading Using Oct Images

To determine the feasibility of automatic Lens Opacities Classification System (LOCS) III grading of human cataractous lenses using software analysis of intraoperative optical coherence tomography (OCT) scans images by leveraging novel machine learning approaches.

Dardenne Eye Hospital (Bonn, Germany)

Retrospective analysis of LOCS III gradings of cataract surgery patient eyes graded by physicians and of OCT images of the same eyes obtained during femtosecond laser assisted cataract surgery using a Ziemer LDV Z8 laser. A tailored Artificial Intelligence software solution based on convolutional neural networks is used to classify the OCT data by extracting local and global scene information including image brightness, textures and structural information. The supervised learning of the AI is performed with a part of the available pre-graded eye data, whereas the validation uses a disjoint, representative dataset of eyes.

We evaluated more than 1000 data sets incl. OCT B-Scan images of the whole lens thickness, from at least 2 perpendicular scans in the temporal-nasal and inferior-superior axes. Correlations of software predictions of LOCS III grades NO, C and P with actual pre-operatively determined LOCS III grades are presented. Different neural network approaches are discussed together with their feasibility for accurate grade prediction. We present the challenges of our novel approach as well as an outlook on future prospects of automated cataract grading.

We present the feasibility of artificial intelligence based prediction of LOCS III grading on anterior segment OCT images. Our novel approach provides a tool for supporting the LOCS III grading of human cataractous lenses and therefore facilitating the clinical analysis.