Joint Supervised and Self-supervised Learning for MRI Reconstruction

George Yiasemis, Nikita Moriakov, Clara I. Sánchez, Jan-Jakob Sonke, Jonas Teuwen
Proceedings of The 8th International Conference on Medical Imaging with Deep Learning, PMLR 301:1771-1794, 2026.

Abstract

Magnetic Resonance Imaging (MRI) is a crucial modality but, its inherently slow acquisition process poses challenges in obtaining fully-sampled $k$-space data under motion. The lack of fully-sampled acquisitions, serving as ground truths, complicates the training of deep learning (DL) algorithms in a supervised manner. To address this limitation, self-supervised learning (SSL) methods have emerged as a viable alternative, leveraging available subsampled $k$-space data to train neural networks for MRI reconstruction. Nevertheless, these approaches often fall short when compared to supervised learning (SL). We propose Joint Supervised and Self-supervised Learning (JSSL), a novel training approach for DL-based MRI reconstruction algorithms aimed at enhancing reconstruction quality in cases where target datasets containing fully-sampled $k$-space measurements are unavailable. JSSL operates by simultaneously training a model in a SSL setting, using subsampled data from the target dataset(s), and in a SL manner, utilizing proxy datasets with fully-sampled $k$-space data. We demonstrate JSSLś efficacy using two distinct combinations of target and proxy data. Quantitative and qualitative results showcase substantial improvements over conventional SSL methods. Furthermore, we provide r̈ule-of-thumb\"{guidelines} for training MRI reconstruction models. Our code is available at https://github.com/NKI-AI/direct.

Cite this Paper

BibTeX
@InProceedings{pmlr-v301-yiasemis26a, title = {Joint Supervised and Self-supervised Learning for MRI Reconstruction}, author = {Yiasemis, George and Moriakov, Nikita and S\'anchez, Clara I. and Sonke, Jan-Jakob and Teuwen, Jonas}, booktitle = {Proceedings of The 8th International Conference on Medical Imaging with Deep Learning}, pages = {1771--1794}, year = {2026}, editor = {Tasdizen, Tolga and Elhabian, Shireen and Summers, Ronald and Chen, Chen and Koch, Lisa and Zhuang, Yan}, volume = {301}, series = {Proceedings of Machine Learning Research}, month = {09--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v301/main/assets/yiasemis26a/yiasemis26a.pdf}, url = {https://proceedings.mlr.press/v301/yiasemis26a.html}, abstract = {Magnetic Resonance Imaging (MRI) is a crucial modality but, its inherently slow acquisition process poses challenges in obtaining fully-sampled $k$-space data under motion. The lack of fully-sampled acquisitions, serving as ground truths, complicates the training of deep learning (DL) algorithms in a supervised manner. To address this limitation, self-supervised learning (SSL) methods have emerged as a viable alternative, leveraging available subsampled $k$-space data to train neural networks for MRI reconstruction. Nevertheless, these approaches often fall short when compared to supervised learning (SL). We propose Joint Supervised and Self-supervised Learning (JSSL), a novel training approach for DL-based MRI reconstruction algorithms aimed at enhancing reconstruction quality in cases where target datasets containing fully-sampled $k$-space measurements are unavailable. JSSL operates by simultaneously training a model in a SSL setting, using subsampled data from the target dataset(s), and in a SL manner, utilizing proxy datasets with fully-sampled $k$-space data. We demonstrate JSSLś efficacy using two distinct combinations of target and proxy data. Quantitative and qualitative results showcase substantial improvements over conventional SSL methods. Furthermore, we provide r̈ule-of-thumb\"{guidelines} for training MRI reconstruction models. Our code is available at https://github.com/NKI-AI/direct.} }
Endnote
%0 Conference Paper %T Joint Supervised and Self-supervised Learning for MRI Reconstruction %A George Yiasemis %A Nikita Moriakov %A Clara I. Sánchez %A Jan-Jakob Sonke %A Jonas Teuwen %B Proceedings of The 8th International Conference on Medical Imaging with Deep Learning %C Proceedings of Machine Learning Research %D 2026 %E Tolga Tasdizen %E Shireen Elhabian %E Ronald Summers %E Chen Chen %E Lisa Koch %E Yan Zhuang %F pmlr-v301-yiasemis26a %I PMLR %P 1771--1794 %U https://proceedings.mlr.press/v301/yiasemis26a.html %V 301 %X Magnetic Resonance Imaging (MRI) is a crucial modality but, its inherently slow acquisition process poses challenges in obtaining fully-sampled $k$-space data under motion. The lack of fully-sampled acquisitions, serving as ground truths, complicates the training of deep learning (DL) algorithms in a supervised manner. To address this limitation, self-supervised learning (SSL) methods have emerged as a viable alternative, leveraging available subsampled $k$-space data to train neural networks for MRI reconstruction. Nevertheless, these approaches often fall short when compared to supervised learning (SL). We propose Joint Supervised and Self-supervised Learning (JSSL), a novel training approach for DL-based MRI reconstruction algorithms aimed at enhancing reconstruction quality in cases where target datasets containing fully-sampled $k$-space measurements are unavailable. JSSL operates by simultaneously training a model in a SSL setting, using subsampled data from the target dataset(s), and in a SL manner, utilizing proxy datasets with fully-sampled $k$-space data. We demonstrate JSSLś efficacy using two distinct combinations of target and proxy data. Quantitative and qualitative results showcase substantial improvements over conventional SSL methods. Furthermore, we provide r̈ule-of-thumb\"{guidelines} for training MRI reconstruction models. Our code is available at https://github.com/NKI-AI/direct.
APA
Yiasemis, G., Moriakov, N., Sánchez, C.I., Sonke, J. & Teuwen, J.. (2026). Joint Supervised and Self-supervised Learning for MRI Reconstruction. Proceedings of The 8th International Conference on Medical Imaging with Deep Learning, in Proceedings of Machine Learning Research 301:1771-1794 Available from https://proceedings.mlr.press/v301/yiasemis26a.html.

Related Material