Abstract
Deep learning models for brain tumor MRI classification reach near-saturated accuracy, yet high performance does not guarantee clinically meaningful representations. We propose a model-agnostic, retraining-free representation audit testing diagnostic-structure, radiomic-correspondence, and spatial-stability hypotheses, finding subtype-dependent attribution stability between focal and infiltrative tumors on a ResNet-50 audit.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MLCN_2026_035.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=ta8GZDjbrr
BibTex
@InProceedings{LaiBo_Do_MICCAISAT2026,
author = { Lai, Bo-Wei AND Chen, Li-Fen},
title = { { Do CNNs Learn Clinically Meaningful Imaging Representations? A Portable Multi-level Representation Audit for Brain Tumor MRI Classification } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17255},
month = {pending},
page = {pending}
}