Primitive Representation Learning for Unsupervised Dynamic Contrast-Enhanced MRI Reconstruction

Forum topic · 小凯 · 2026-08-20

Summary

This paper (arXiv:2608.18055) proposes a multi-dimensional, primitive-based framework for unsupervised reconstruction of dynamic contrast-enhanced (DCE) MRI. Reliable quantitative analysis of DCE-MRI requires high-quality spatiotemporal reconstructions at high undersampling rates. Prior scan-specific methods using Gaussian and Gabor primitives achieved promising results without large training datasets, but did not handle the additional dynamic contrast dimension. The proposed framework disentangles underlying anatomy, dynamic contrast enhancement, and residual motion into separate temporal basis functions, enabling a geometric interpretation of the learned representation. Experiments show performance competitive with conventional reconstruction methods in both reconstruction quality and accuracy of extracted aortic and kidney enhancement curves. The modular hierarchical design naturally extends to additional dynamic factors and higher acceleration rates. Code is available at https://github.com/compai-lab/2026-GaborDCE-spieker.

Overview

Field: Computer Vision / Medical Imaging Authors: Veronika Spieker, Wenqi Huang, Cemre Ariyurek, Liam Timms, Daniel Rueckert, Onur Afacan, Julia A. Schnabel, Sila Kurugol Published: 2026-08-18 arXiv: 2608.18055

Abstract

Reliable quantitative analysis of dynamic contrast-enhanced MRI requires high-quality spatiotemporal reconstructions at high undersampling rates. Scan-specific reconstructions using Gaussian and Gabor primitives have shown promising results without the need for large training datasets, but have not addressed the additional dimension of dynamic contrast.

We propose a multi-dimensional, primitive-based framework for dynamic contrast-enhanced MRI reconstruction that disentangles the underlying anatomy, the dynamic contrast enhancement, and residual motion into separate temporal basis functions, thereby enabling a geometrical interpretation of the representation.

We show that this architecture achieves performance competitive with conventional reconstruction methods, both in reconstruction quality and in the accuracy of extracted aortic and kidney enhancement curves. The modular hierarchical design naturally extends to additional dynamic factors and higher acceleration rates.

Code

https://github.com/compai-lab/2026-GaborDCE-spieker