Regularized multitask learning with auxiliary target for foveal avascular zone segmentation in fluorescein angiography

DOMINIK HOFER; JOSÉ IGNACIO ORLANDO; PHILLIP SEEBÖCK; BIANCA S. GERENDAS; URSULA SCHMIDT-ERFURTH

Viena

Jornada; Medical Imaging Cluster (MIC) Festival (2019); 2019

Medical University of Vienna

The Foveal Avascular Zone (FAZ) is a vessel-free area in the center ofthe retina that is used for diagnosis of eye-related diseases such asdiabetic retinopathy. Segmentation of the FAZ in Fluorescein Angiography(FA) plays an important role in clinical decision making, as the shapeand area of the FAZ are used as biomarkers for detecting and grading theseverity of pathologies. In this paper we implement a multi-decoder U-shaped based architecture for FAZ segmentation in FA. Our multitasklearning approach takes advantage of an auxiliary prediction task thataids to regularize the results, makes the result more smooth and compactand deletes possible outliers.