Weight Space Correlation Analysis: Quantifying Feature Utilization in Deep Learning Models

Chun Kit Wong, Paraskevas Pegios, Nina Weng, Emilie Pi Fogtmann Sejer, Martin Grønnebæk Tolsgaard, Anders Nymark Christensen, Aasa Feragen
Proceedings of The 9th International Conference on Medical Imaging with Deep Learning, PMLR 315:2711-2737, 2026.

Abstract

Deep learning models in medical imaging are susceptible to shortcut learning, relying on confounding metadata (e.g. scanner model) that is often encoded in image embeddings. The crucial question is whether the model actively utilizes this encoded information for its final prediction. We introduce Weight Space Correlation analysis, an interpretable methodology that quantifies feature utilization by measuring the alignment between the classification heads of a primary clinical task and auxiliary metadata tasks. We first validate our method by successfully detecting artificially induced shortcut learning. We then apply it to probe the feature utilization of an SA-SonoNet model trained for Spontaneous Preterm Birth (sPTB) prediction. Our analysis confirmed that while the embeddings contain substantial metadata, the sPTB classifier’s weight vectors were highly correlated with clinically relevant factors (e.g. cervical length) but decoupled from clinically irrelevant acquisition factors (e.g. scanner). Our methodology provides a tool for verifying model trustworthiness, by inspecting whether it utilizes features unrelated to the genuine clinical signal.

Cite this Paper

BibTeX
@InProceedings{pmlr-v315-wong26a, title = {Weight Space Correlation Analysis: Quantifying Feature Utilization in Deep Learning Models}, author = {Wong, Chun Kit and Pegios, Paraskevas and Weng, Nina and Sejer, Emilie Pi Fogtmann and Tolsgaard, Martin Gr{\o}nneb{\ae}k and Christensen, Anders Nymark and Feragen, Aasa}, booktitle = {Proceedings of The 9th International Conference on Medical Imaging with Deep Learning}, pages = {2711--2737}, year = {2026}, editor = {Huo, Yuankai and Gao, Mingchen and Kuo, Chang-Fu and Jin, Yueming and Deng, Ruining}, volume = {315}, series = {Proceedings of Machine Learning Research}, month = {08--10 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v315/main/assets/wong26a/wong26a.pdf}, url = {https://proceedings.mlr.press/v315/wong26a.html}, abstract = {Deep learning models in medical imaging are susceptible to shortcut learning, relying on confounding metadata (e.g. scanner model) that is often encoded in image embeddings. The crucial question is whether the model actively utilizes this encoded information for its final prediction. We introduce Weight Space Correlation analysis, an interpretable methodology that quantifies feature utilization by measuring the alignment between the classification heads of a primary clinical task and auxiliary metadata tasks. We first validate our method by successfully detecting artificially induced shortcut learning. We then apply it to probe the feature utilization of an SA-SonoNet model trained for Spontaneous Preterm Birth (sPTB) prediction. Our analysis confirmed that while the embeddings contain substantial metadata, the sPTB classifier’s weight vectors were highly correlated with clinically relevant factors (e.g. cervical length) but decoupled from clinically irrelevant acquisition factors (e.g. scanner). Our methodology provides a tool for verifying model trustworthiness, by inspecting whether it utilizes features unrelated to the genuine clinical signal.} }
Endnote
%0 Conference Paper %T Weight Space Correlation Analysis: Quantifying Feature Utilization in Deep Learning Models %A Chun Kit Wong %A Paraskevas Pegios %A Nina Weng %A Emilie Pi Fogtmann Sejer %A Martin Grønnebæk Tolsgaard %A Anders Nymark Christensen %A Aasa Feragen %B Proceedings of The 9th International Conference on Medical Imaging with Deep Learning %C Proceedings of Machine Learning Research %D 2026 %E Yuankai Huo %E Mingchen Gao %E Chang-Fu Kuo %E Yueming Jin %E Ruining Deng %F pmlr-v315-wong26a %I PMLR %P 2711--2737 %U https://proceedings.mlr.press/v315/wong26a.html %V 315 %X Deep learning models in medical imaging are susceptible to shortcut learning, relying on confounding metadata (e.g. scanner model) that is often encoded in image embeddings. The crucial question is whether the model actively utilizes this encoded information for its final prediction. We introduce Weight Space Correlation analysis, an interpretable methodology that quantifies feature utilization by measuring the alignment between the classification heads of a primary clinical task and auxiliary metadata tasks. We first validate our method by successfully detecting artificially induced shortcut learning. We then apply it to probe the feature utilization of an SA-SonoNet model trained for Spontaneous Preterm Birth (sPTB) prediction. Our analysis confirmed that while the embeddings contain substantial metadata, the sPTB classifier’s weight vectors were highly correlated with clinically relevant factors (e.g. cervical length) but decoupled from clinically irrelevant acquisition factors (e.g. scanner). Our methodology provides a tool for verifying model trustworthiness, by inspecting whether it utilizes features unrelated to the genuine clinical signal.
APA
Wong, C.K., Pegios, P., Weng, N., Sejer, E.P.F., Tolsgaard, M.G., Christensen, A.N. & Feragen, A.. (2026). Weight Space Correlation Analysis: Quantifying Feature Utilization in Deep Learning Models. Proceedings of The 9th International Conference on Medical Imaging with Deep Learning, in Proceedings of Machine Learning Research 315:2711-2737 Available from https://proceedings.mlr.press/v315/wong26a.html.

Related Material