Deep multi-task learning for nephropathy diagnosis on immunofluorescence images

Yibing Fu, Lai Jiang, Sai Pan, Pu Chen, Xiaofei Wang, Ning Dai, Xiangmei Chen, Mai Xu

Abstract: Highlights•We focus on scenarios of nephropathy diagnosis on blurred immunofluorescence images, which leads to performance degradation.•We establish two open-source databases including 6,521 immunofluorescence images with synthetic or real-world blurs.•We thoroughly analyze our databases for investigating the impact of image quality on disease diagnosis.•We propose a hierarchical multi-task learning method, bridging the gap between low- and high-level medical tasks.