Hide-and-Seek Attribution: Weakly Supervised Segmentation of Vertebral Metastases in CT

Matan Atad, Alexander W. Marka, Lisa Steinhelfer, Anna Curto-Vilalta, Yannik Leonhardt, Sarah C. Foreman, Anna-Sophia Walburga Dietrich, Robert Graf, Alexandra S. Gersing, Bjoern Menze, Daniel Rueckert, Jan S. Kirschke, Hendrik Moeller
Proceedings of The 9th International Conference on Medical Imaging with Deep Learning, PMLR 315:960-988, 2026.

Abstract

Accurate segmentation of vertebral metastasis in CT is clinically important yet difficult to scale, as voxel-level annotations are scarce and both lytic and blastic lesions often resemble benign degenerative changes. We introduce a 2D weakly supervised method trained solely on vertebra-level healthy/malignant labels, without any lesion masks. The method combines a Diffusion Autoencoder (DAE) that produces a classifier-guided healthy edit of each vertebra with pixel-wise difference maps that propose suspect candidate lesions. To determine which regions truly reflect malignancy, we introduce Hide-and-Seek Attribution: each candidate is revealed in turn while all others are hidden, the edited image is projected back to the data manifold by the DAE, and a latent-space classifier quantifies the isolated malignant contribution of that component. High-scoring regions form the final lytic or blastic segmentation. On held-out radiologist annotations, we achieve strong blastic/lytic performance despite no mask supervision (F1: 0.91/0.85; Dice: 0.87/0.78), exceeding baselines (F1: 0.79/0.67; Dice: 0.74/0.55). These results show that vertebra-level labels can be transformed into reliable lesion masks, demonstrating that generative editing combined with selective occlusion supports accurate weakly supervised segmentation in CT.

Cite this Paper

BibTeX
@InProceedings{pmlr-v315-atad26a, title = {Hide-and-Seek Attribution: Weakly Supervised Segmentation of Vertebral Metastases in CT}, author = {Atad, Matan and Marka, Alexander W. and Steinhelfer, Lisa and Curto-Vilalta, Anna and Leonhardt, Yannik and Foreman, Sarah C. and Dietrich, Anna-Sophia Walburga and Graf, Robert and Gersing, Alexandra S. and Menze, Bjoern and Rueckert, Daniel and Kirschke, Jan S. and Moeller, Hendrik}, booktitle = {Proceedings of The 9th International Conference on Medical Imaging with Deep Learning}, pages = {960--988}, 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/atad26a/atad26a.pdf}, url = {https://proceedings.mlr.press/v315/atad26a.html}, abstract = {Accurate segmentation of vertebral metastasis in CT is clinically important yet difficult to scale, as voxel-level annotations are scarce and both lytic and blastic lesions often resemble benign degenerative changes. We introduce a 2D weakly supervised method trained solely on vertebra-level healthy/malignant labels, without any lesion masks. The method combines a Diffusion Autoencoder (DAE) that produces a classifier-guided healthy edit of each vertebra with pixel-wise difference maps that propose suspect candidate lesions. To determine which regions truly reflect malignancy, we introduce Hide-and-Seek Attribution: each candidate is revealed in turn while all others are hidden, the edited image is projected back to the data manifold by the DAE, and a latent-space classifier quantifies the isolated malignant contribution of that component. High-scoring regions form the final lytic or blastic segmentation. On held-out radiologist annotations, we achieve strong blastic/lytic performance despite no mask supervision (F1: 0.91/0.85; Dice: 0.87/0.78), exceeding baselines (F1: 0.79/0.67; Dice: 0.74/0.55). These results show that vertebra-level labels can be transformed into reliable lesion masks, demonstrating that generative editing combined with selective occlusion supports accurate weakly supervised segmentation in CT.} }
Endnote
%0 Conference Paper %T Hide-and-Seek Attribution: Weakly Supervised Segmentation of Vertebral Metastases in CT %A Matan Atad %A Alexander W. Marka %A Lisa Steinhelfer %A Anna Curto-Vilalta %A Yannik Leonhardt %A Sarah C. Foreman %A Anna-Sophia Walburga Dietrich %A Robert Graf %A Alexandra S. Gersing %A Bjoern Menze %A Daniel Rueckert %A Jan S. Kirschke %A Hendrik Moeller %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-atad26a %I PMLR %P 960--988 %U https://proceedings.mlr.press/v315/atad26a.html %V 315 %X Accurate segmentation of vertebral metastasis in CT is clinically important yet difficult to scale, as voxel-level annotations are scarce and both lytic and blastic lesions often resemble benign degenerative changes. We introduce a 2D weakly supervised method trained solely on vertebra-level healthy/malignant labels, without any lesion masks. The method combines a Diffusion Autoencoder (DAE) that produces a classifier-guided healthy edit of each vertebra with pixel-wise difference maps that propose suspect candidate lesions. To determine which regions truly reflect malignancy, we introduce Hide-and-Seek Attribution: each candidate is revealed in turn while all others are hidden, the edited image is projected back to the data manifold by the DAE, and a latent-space classifier quantifies the isolated malignant contribution of that component. High-scoring regions form the final lytic or blastic segmentation. On held-out radiologist annotations, we achieve strong blastic/lytic performance despite no mask supervision (F1: 0.91/0.85; Dice: 0.87/0.78), exceeding baselines (F1: 0.79/0.67; Dice: 0.74/0.55). These results show that vertebra-level labels can be transformed into reliable lesion masks, demonstrating that generative editing combined with selective occlusion supports accurate weakly supervised segmentation in CT.
APA
Atad, M., Marka, A.W., Steinhelfer, L., Curto-Vilalta, A., Leonhardt, Y., Foreman, S.C., Dietrich, A.W., Graf, R., Gersing, A.S., Menze, B., Rueckert, D., Kirschke, J.S. & Moeller, H.. (2026). Hide-and-Seek Attribution: Weakly Supervised Segmentation of Vertebral Metastases in CT. Proceedings of The 9th International Conference on Medical Imaging with Deep Learning, in Proceedings of Machine Learning Research 315:960-988 Available from https://proceedings.mlr.press/v315/atad26a.html.

Related Material