Weakly Supervised Learning for Subcutaneous Edema Segmentation of Abdominal CT

Authors

• S Bhadra
• J Liu
• RM Summers

Abstract

Anasarca is described as excessive accumulation of interstitial fluids within the subcutaneous adipose tissue, causing generalized edema. It mainly occurs due to organ dysfunction, such as heart, kidney or liver failure. Volumetric assessment of edema due to anasarca can help monitor the progression of these diseases. The ability to measure edema non-invasively by automatic segmentation from abdominal CT scans may be of clinical importance. The current state-of-the-art method for edema segmentation using intensity priors is susceptible to false positives or under-segmentation errors. The application of modern supervised deep learning methods for 3D edema segmentation is limited due to challenges in manual annotation of edema. In the absence of accurate 3D annotations of edema, we propose a weakly supervised learning method that uses edema segmentations produced by intensity priors as pseudo-labels, along with pseudo-labels of muscle, subcutaneous and visceral adipose tissues for context, to produce more refined segmentations. Our method employs nnU-Net as the deep learning segmentation backbone in multiple stages to produce the final edema segmentation. The proposed method improved the average Dice Similarity Coefficient and relative volume difference of edema by 4-5 % (p<0.05) compared to the intensity prior method. Qualitatively, we observed that weak supervision with multi-class pseudo-labels significantly mitigated the false positives and under-segmentation errors produced by the unsupervised intensity prior method. The results demonstrate the potential of weakly supervised learning in improved quantification of edema for monitoring anasarca.

Scientific Focus Area: Biomedical Engineering and Biophysics

This page was last updated on Tuesday, August 6, 2024