Abstract

Precise thickness measurements of retinallayers are crucial to decide whether the subject re-quires subsequent treatment. As optical coherencetomography (OCT) is becoming a standard imagingmethod in hospitals, the amount of retinal scans in-creases rapidly, automated segmentation algorithmsare getting deployed, and methods to assess theirperformance are in demand.In this work we propose a semi-supervised frame-work to detect incorrectly segmented OCT retinascans: ground-truth segmentations are (1) embed-ded in 2D feature space and (2) used to train an out-lier scoring function and the corresponding decisionboundary.We evaluate a selection of five outlier detectionmethods and find the results to be a promising start-ing point to address the given problem. While thiswork and results are centred around one concretesegmentation algorithm we sketch the possibilities ofhow the framework can be generalized for more re-cent or more precise segmentation methods.

Reference

Hladuvka, J., & Renner, V. (2020). Towards Identification of Incorrectly Segmented OCT Scans. In Proceedings of the Joint Austrian Computer Vision and Robotics Workshop 2020 (p. 7). https://doi.org/10.3217/978-3-85125-752-6-36