Multiclass segmentation as multitask learning for drusen segmentation in retinal optical coherence tomography

RHONA ASGARI; JOSÉ IGNACIO ORLANDO; SEBASTIAN WALDSTEIN; FERDINAND SCHLANITZ; MAGDALENA BARATSITS; URSULA SCHMIDT-ERFURTH; HRVOJE BOGUNOVIC

Vienna

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

Medical University of Vienna

Automated drusen segmentation in retinal optical coherencetomography (OCT) scans is relevant for understanding age-related mac-ular degeneration (AMD) risk and progression. This task is usually per-formed by segmenting the top/bottom anatomical interfaces that definedrusen, the outer boundary of the retinal pigment epithelium (OBRPE)and the Bruch?s membrane (BM), respectively. In this paper we pro-pose a novel multi-decoder architecture that tackles drusen segmenta-tion as a multitask problem. Instead of training a multiclass model forOBRPE/BM segmentation, we use one decoder per target class and anextra one aiming for the area between the layers. We also introduce con-nections between each class-specific branch and the additional decoderto increase the regularization effect of this surrogate task. We validatedour approach on private/public data sets with 166 early/intermediateAMD Spectralis, and 200 AMD and control Bioptigen OCT volumes,respectively. Our method consistently outperformed several baselines inboth layer and drusen segmentation evaluations.