TY - JOUR
AU - Chatterjee, Soumick
AU - Yassin, Hadya
AU - Dubost, Florian
AU - Nürnberger, Andreas
AU - Speck, Oliver
TI - Weakly-supervised segmentation using inherently-explainable classification models and their application to brain tumour classification
JO - Neurocomputing
VL - 682
SN - 0925-2312
CY - Amsterdam
PB - Elsevier
M1 - DZNE-2026-00388
SP - 133460
PY - 2026
AB - Deep learning has demonstrated significant potential in medical imaging; however, the opacity of “black-box” models hinders clinical trust, while segmentation tasks typically necessitate laborious, hard-to-obtain pixel-wise annotations. To address these challenges simultaneously, this paper introduces a framework for three inherently explainable classifiers (GP-UNet, GP-ShuffleUNet, and GP-ReconResNet). By integrating a global pooling mechanism, these networks generate localisation heatmaps that directly influence classification decisions, offering inherent interpretability without relying on potentially unreliable post-hoc methods. These heatmaps are subsequently thresholded to achieve weakly-supervised segmentation, requiring only image-level classification labels for training. Validated on two datasets for multi-class brain tumour classification, the proposed models achieved a peak F1-score of 0.93. For the weakly-supervised segmentation task, a median Dice score of 0.728 (95
LB - PUB:(DE-HGF)16
DO - DOI:10.1016/j.neucom.2026.133460
UR - https://pub.dzne.de/record/286092
ER -