[论文解读] Self-Supervised Deep Active Accelerated MRI

Kyong Hwan Jin, Michaël Unser+1 more|arXiv (Cornell University)|Jan 14, 2019

Medical Imaging Techniques and Applications参考文献 33被引用 47

一句话总结

作者在加速 MRI 中联合学习采样与重建，使用由蒙特卡罗树搜索引导的自监督框架，训练两网络（ReconNet 与 SampleNet）在固定采样预算下提高重建质量。

ABSTRACT

We propose to simultaneously learn to sample and reconstruct magnetic resonance images (MRI) to maximize the reconstruction quality given a limited sample budget, in a self-supervised setup. Unlike existing deep methods that focus only on reconstructing given data, thus being passive, we go beyond the current state of the art by considering both the data acquisition and the reconstruction process within a single deep-learning framework. As our network learns to acquire data, the network is active in nature. In order to do so, we simultaneously train two neural networks, one dedicated to reconstruction and the other to progressive sampling, each with an automatically generated supervision signal that links them together. The two supervision signals are created through Monte Carlo tree search (MCTS). MCTS returns a better sampling pattern than what the current sampling network can give and, thus, a better final reconstruction. The sampling network is trained to mimic the MCTS results using the previous sampling network, thus being enhanced. The reconstruction network is trained to give the highest reconstruction quality, given the MCTS sampling pattern. Through this framework, we are able to train the two networks without providing any direct supervision on sampling.

研究动机与目标

• 将加速 MRI 作为一个联合采样与重建的问题来 motivate。
• Develop a self-supervised framework that learns both data acquisition and image reconstruction.
• Train two interconnected networks (ReconNet and SampleNet) to optimize sampling strategy and image quality without direct sampling supervision.
• Demonstrate that integrating sampling design with reconstruction yields superior results over fixed-pattern methods.

提出的方法

• 提出两种神经网络：ReconNet 用于高质量重建，SampleNet 用于提出下一个采样位置。
• Use progressive, one-sample-at-a-time sampling and link networks via self-generated supervision.
• Employ Monte Carlo tree search (MCTS) to generate sampling patterns that improve reconstruction, guiding both networks.
• Train ReconNet with PSNR-based supervision using MCTS-sampled data; train SampleNet to mimic MCTS policies using cross-entropy loss.
• Iteratively update networks through self-supervised rounds with experience replay to stabilize learning.

实验结果

研究问题

• RQ1 Can sampling pattern design be learned jointly with reconstruction to maximize MRI quality under a sampling budget?
• RQ2 Does self-supervision via MCTS provide effective guidance to jointly train sampling and reconstruction networks?
• RQ3 How does the proposed framework compare to fixed sampling patterns and reconstruction pipelines in accelerated MRI?
• RQ4 What is the impact of progressive sampling and MCTS-driven supervision on reconstruction metrics like PSNR?

主要发现

• The method achieves higher PSNR than a prior approach (33.52 dB vs 27.73 dB in the cited baseline).
• Two networks trained jointly—ReconNet and SampleNet—outperform fixed-pattern sampling schemes.
• MCTS-generated sampling patterns yield better reconstructions than the sampling network alone.
• Training relies on self-generated supervision without direct sampling labels, enabling end-to-end optimization of acquisition and reconstruction.
• The framework demonstrates performance gains on cardiac and knee MRI datasets compared to baselines such as VDS+TV and LCS+TV.