Branch Learning in MRI: More Data, More Models, More Training

TL;DR

This paper explores advanced data augmentation and model scaling techniques for multicontrast cardiac MRI reconstruction, demonstrating their effects on performance, generalization, and efficiency across diverse datasets and models.

Contribution

It introduces physics-consistent augmentation and prompt-based capacity scaling methods, showing their benefits and limitations in improving MRI reconstruction and model efficiency.

Findings

K-space motion-plus-noise augmentation improves small dataset reconstruction.

VQPrompt yields modest, consistent gains with low overhead.

Moero enhances performance and maintains generalization despite overfitting.

Abstract

We investigated two complementary strategies for multicontrast cardiac MR reconstruction: physics-consistent data-space augmentation (DualSpaceCMR) and parameter-efficient capacity scaling via VQPrompt and Moero. DualSpaceCMR couples image-level transforms with kspace noise and motion simulations while preserving forwardmodel consistency. VQPrompt adds a lightweight bottleneck prompt; Moero embeds a sparse mixture of experts within a deep unrolled network with histogram-based routing. In the multivendor, multisite CMRxRecon25 benchmark, we evaluate fewshot and out-of-distribution generalization. On small datasets, k-space motion-plus-noise improves reconstruction; on the large benchmark it degrades performance, revealing sensitivity to augmentation ratio and schedule. VQPrompt produces modest and consistent gains with negligible memory overhead. Moero continues to improve after early plateaus and maintains baseline-like fewshot and out-of-distribution behavior despite mild overfitting, but sparse routing lowers PyTorch throughput and makes wall clock time the main bottleneck. These results motivate scale-aware augmentation and suggest prompt-based capacity scaling as a practical path, while efficiency improvements are crucial for sparse expert models.