Learning Morphological Feature Perturbations for Calibrated Semi-Supervised Segmentation

Mou-Cheng Xu, Yu-Kun Zhou, Chen Jin, Stefano B. Blumberg, Frederick J. Wilson, Marius deGroot, Daniel C. Alexander, Neil P. Oxtoby, Joseph Jacob
Proceedings of The 5th International Conference on Medical Imaging with Deep Learning, PMLR 172:1413-1429, 2022.

Abstract

We propose MisMatch, a novel consistency-driven semi-supervised segmentation framework which produces predictions that are invariant to learnt feature perturbations. MisMatch consists of an encoder and a two-head decoders. One decoder learns positive attention to the foreground regions of interest (RoI) on unlabelled images thereby generating dilated features. The other decoder learns negative attention to the foreground on the same unlabelled images thereby generating eroded features. We then apply a consistency regularisation on the paired predictions. MisMatch outperforms state-of-the-art semi-supervised methods on a CT-based pulmonary vessel segmentation task and a MRI-based brain tumour segmentation task. In addition, we show that the effectiveness of MisMatch comes from better model calibration than its supervised learning counterpart.

Cite this Paper

BibTeX
@InProceedings{pmlr-v172-xu22a, title = {Learning Morphological Feature Perturbations for Calibrated Semi-Supervised Segmentation}, author = {Xu, Mou-Cheng and Zhou, Yu-Kun and Jin, Chen and Blumberg, Stefano B. and Wilson, Frederick J. and deGroot, Marius and Alexander, Daniel C. and Oxtoby, Neil P. and Jacob, Joseph}, booktitle = {Proceedings of The 5th International Conference on Medical Imaging with Deep Learning}, pages = {1413--1429}, year = {2022}, editor = {Konukoglu, Ender and Menze, Bjoern and Venkataraman, Archana and Baumgartner, Christian and Dou, Qi and Albarqouni, Shadi}, volume = {172}, series = {Proceedings of Machine Learning Research}, month = {06--08 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v172/xu22a/xu22a.pdf}, url = {https://proceedings.mlr.press/v172/xu22a.html}, abstract = {We propose MisMatch, a novel consistency-driven semi-supervised segmentation framework which produces predictions that are invariant to learnt feature perturbations. MisMatch consists of an encoder and a two-head decoders. One decoder learns positive attention to the foreground regions of interest (RoI) on unlabelled images thereby generating dilated features. The other decoder learns negative attention to the foreground on the same unlabelled images thereby generating eroded features. We then apply a consistency regularisation on the paired predictions. MisMatch outperforms state-of-the-art semi-supervised methods on a CT-based pulmonary vessel segmentation task and a MRI-based brain tumour segmentation task. In addition, we show that the effectiveness of MisMatch comes from better model calibration than its supervised learning counterpart.} }
Endnote
%0 Conference Paper %T Learning Morphological Feature Perturbations for Calibrated Semi-Supervised Segmentation %A Mou-Cheng Xu %A Yu-Kun Zhou %A Chen Jin %A Stefano B. Blumberg %A Frederick J. Wilson %A Marius deGroot %A Daniel C. Alexander %A Neil P. Oxtoby %A Joseph Jacob %B Proceedings of The 5th International Conference on Medical Imaging with Deep Learning %C Proceedings of Machine Learning Research %D 2022 %E Ender Konukoglu %E Bjoern Menze %E Archana Venkataraman %E Christian Baumgartner %E Qi Dou %E Shadi Albarqouni %F pmlr-v172-xu22a %I PMLR %P 1413--1429 %U https://proceedings.mlr.press/v172/xu22a.html %V 172 %X We propose MisMatch, a novel consistency-driven semi-supervised segmentation framework which produces predictions that are invariant to learnt feature perturbations. MisMatch consists of an encoder and a two-head decoders. One decoder learns positive attention to the foreground regions of interest (RoI) on unlabelled images thereby generating dilated features. The other decoder learns negative attention to the foreground on the same unlabelled images thereby generating eroded features. We then apply a consistency regularisation on the paired predictions. MisMatch outperforms state-of-the-art semi-supervised methods on a CT-based pulmonary vessel segmentation task and a MRI-based brain tumour segmentation task. In addition, we show that the effectiveness of MisMatch comes from better model calibration than its supervised learning counterpart.
APA
Xu, M., Zhou, Y., Jin, C., Blumberg, S.B., Wilson, F.J., deGroot, M., Alexander, D.C., Oxtoby, N.P. & Jacob, J.. (2022). Learning Morphological Feature Perturbations for Calibrated Semi-Supervised Segmentation. Proceedings of The 5th International Conference on Medical Imaging with Deep Learning, in Proceedings of Machine Learning Research 172:1413-1429 Available from https://proceedings.mlr.press/v172/xu22a.html.

Related Material