AUTHOR=Kong Zishang , He Min , Luo Qianjiang , Huang Xiansong , Wei Pengxu , Cheng Yalu , Chen Luyang , Liang Yongsheng , Lu Yanchang , Li Xi , Chen Jie 

TITLE=Multi-Task Classification and Segmentation for Explicable Capsule Endoscopy Diagnostics

JOURNAL=Frontiers in Molecular Biosciences

VOLUME=Volume 8 - 2021

YEAR=2021

URL=https://www.frontiersin.org/journals/molecular-biosciences/articles/10.3389/fmolb.2021.614277

DOI=10.3389/fmolb.2021.614277

ISSN=2296-889X

ABSTRACT=Capsule Endoscopy is a leading diagnostic tool for small bowel lesions which is faced with
challenges, like time consuming interpreting and harsh optical environment inside small intestine.
Specialists unavoidably wasted lots of time on searching high clearness degree image for
accurate diagnostic. However, current clearness degree classification methods are either based
on traditional attributes or unexplainable deep neural network. In this paper, we propose a
multi-task framework, called Multi-Task Classification and Segmentation Network(MTCSN), to
achieve a joint learning of clearness degree(CD) and tissue semantic segmentation(TSS) for
the first time. In MTCSN, CD helps to generate better refined TSS, while TSS provides an
explicable semantic map to better classify the CD. In addition, we present a new benchmark,
named Capsule-Endoscopy Crohn’s Disease dataset, which introduces the challenges faced
in real world including motion blur, excreta occlusion, refection and various complex alimentary
scenes that are widely existed in endoscopy examination. Extensive experiments and ablation
studies report the significant performance gains of MTCSN over state-of-the-art methods.