AUTHOR=Vengalil Sunil Kumar , Krishnamurthy Bharath , Sinha Neelam 

TITLE=Simultaneous segmentation of multiple structures in fundal images using multi-tasking deep neural networks

JOURNAL=Frontiers in Signal Processing

VOLUME=Volume 2 - 2022

YEAR=2023

URL=https://www.frontiersin.org/journals/signal-processing/articles/10.3389/frsip.2022.936875

DOI=10.3389/frsip.2022.936875

ISSN=2673-8198

ABSTRACT=Fundal imaging is the most commonly used non-invasive technique for early detection of many retinal diseases like diabetic retinopathy. An initial step in automatic processing of fundal images for detecting diseases is to identify the various landmark regions like Optic Disc (OD), Blood Vessels (BV) and Macula. In addition to these, various abnormalities like Exudates that help in pathological analysis are also visible in fundal images. In this work, we propose a multi-tasking deep learning architecture for segmenting Optic Disc, Blood Vessels, Macula and Exudates simultaneously. Our experimental results on publicly available datasets show that simultaneous segmentation of all these structures results in significant improvement in the performance. For segmentation performance on Blood Vessels, we got Dice scores of 80.31%, 81.66% and 80.45% on the datasets DRIVE, HRF, and CHASE\_DB respectively. On Exudates, we got a Dice score of 65% on the IDRiD dataset when trained in combination with Optic Disc (OD) and Blood Vessels using multi-tasking loss function, whereas the Dice score when trained individually is 50%. On the DRIVE dataset our accuracy of our Blood Vessels segmentation results are 1.39% higher than one of the recently reported studies.