Abstract
The interpretation of deep learning decisions in ovarian tumour diagnosis
is a fundamental requirement for clinical adoption and trust. While
convolutional neural networks achieve high accuracy in ultrasound
classification, their decision making processes remain opaque,
necessitating post classification explainability. ExBale is proposed
as a composite alignment metric that quantifies how well explainability
heatmaps correspond to expert annotated tumour segmentation masks by
combining segmentation coverage, CAM containment, and weighted
activation intensity via the geometric mean. The framework is evaluated
across five backbone architectures and six explainability methods using
the MMOTU dataset. Activation based class activation mapping methods
demonstrate statistical superiority over gradient based methods
(Wilcoxon signed rank, $p < 10^{-42}$ across all backbones), with
Swin Transformer and Eigen CAM achieving the highest global alignment
score of 47.85\%, placing it at 23.1\% of the range between a random
noise baseline (32.22\%) and a perfect alignment ceiling (100\%).
Spatial alignment metrics serve as a secondary evaluation
signal for explainability quality; their utility as a standalone
triage criterion is limited by modest Youden J indices
(range 0.001 to 0.138) and the absence of clinical outcome validation.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MIRASOL_009.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=pbbpgWFRgO
BibTex
@InProceedings{KamMin_ExBale_MICCAISAT2026,
author = { Kamel, Mina AND Nour, Mahmoud AND Awad, Mina A. AND Abdelhamed, Mohamed A. AND Selim, Sahar},
title = { { ExBale: Quantitative Evaluation of Explainability Alignment in Ovarian Ultrasound Imaging } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17264},
month = {pending},
page = {pending}
}