Leveraging Probabilistic Segmentation Models for Improved Glaucoma Diagnosis: A Clinical Pipeline Approach

Anna M. Wundram, Paul Fischer, Stephan Wunderlich, Hanna Faber, Lisa M. Koch, Philipp Berens, Christian F. Baumgartner

Keywords: machine learning, segmentation, uncertainty quantification, optic cup and disc segmentation, glaucoma

Abstract: The accurate segmentation of the optic cup and disc in fundus images is essential for diagnostic processes such as glaucoma detection. The inherent ambiguity in locating these structures often poses a significant challenge, leading to potential misdiagnosis. To model such ambiguities, numerous probabilistic segmentation models have been proposed. In this paper, we investigate the integration of these probabilistic segmentation models into a multistage pipeline closely resembling clinical practice. Our findings indicate that leveraging the uncertainties provided by these models substantially enhances the quality of glaucoma diagnosis compared to relying on a single segmentation only.

Submission Number: 26