Feature Reliability and Uncertainty-Aware Fuzzy Learning for COVID-19 Lesion Segmentation

DOI

10.2991/978-94-6239-713-2_8How to use a DOI?

Keywords

COVID-19 lesion segmentation; fuzzy reasoning; medical image segmentation; U-Net; uncertainty modelling

Abstract

There is still the difficult task of accurate delineation of the COVID-19 infection regions in the CT-based scan of the chest, since the lesions have blurred edges, the contrast between infected and noninfected tissues is low, and the lesion appearance is highly variable. Though encoder-decoder-based architectures like U-Net and its variations have been shown to perform well, they are generally not capable of coping with uncertainty and noise around lesion edges, producing segmentation output in fragments or with errors. In order to solve these issues, we introduce AFR-Net, an adaptive fuzzy reasoning segmentation network that explicitly represents feature reliability and boundary uncertainty in a single deep learning model. The given approach is developed by the fuzzy feature reliability module that is aimed at estimating the reliability of encoder features based on the fuzzy inference rules, so that unreliable or noisy features can be blocked during feature fusion. Moreover, the decoder has a fuzzy boundary consistency module that further refines the lesion boundaries based on the reasoning of the local gradient variations and regional homogeneity. All of these fuzzy reasoning elements cooperate to steer the segmentation process, enabling the network to capture the ambiguous regions more appropriately and come up with more consistent lesion boundaries. The proposed framework is evaluated on the MosMedData COVID-19 CT dataset, which contains 1110 chest CT scans with pixel-level infection annotations, and compared with competitive baseline models, including U-Net, U-Net + +, and ResUNet. The results of the experiment demonstrate that AFR-Net has more precise and consistent segmentation performance, indicating its efficiency as an interpretable and uncertainty-sensitive model to segment COVID-19 lesions.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Dhiraj Kumar Raut
AU  - Arla Gopala Krishna
AU  - Sowkuntla Pandu
PY  - 2026
DA  - 2026/06/25
TI  - Feature Reliability and Uncertainty-Aware Fuzzy Learning for COVID-19 Lesion Segmentation
BT  - Proceedings of the International Conference on Advances in Computing Technology and Artificial Intelligence (COMPUTATIA 2026)
PB  - Atlantis Press
SP  - 111
EP  - 124
SN  - 2589-4919
UR  - https://doi.org/10.2991/978-94-6239-713-2_8
DO  - 10.2991/978-94-6239-713-2_8
ID  - Raut2026
ER  -