Parallel Magnetic Resonance Imaging Reconstruction Based On One-shot Learning

Magnetic resonance imaging is a technique that reconstructs images by collecting signals from atomic nuclei resonating in magnetic fields.It serves as an important tool for clinical diagnosis.However,it suffers from a long acquisition time.The utilization of deep learning,especially deep generative models,offers aggressive acceleration and better reconstruction in magnetic resonance imaging.Nevertheless,learning the data distribution as prior knowledge and reconstructing the image from limited data remains challenging.In this work,this article propose a novel Hankel-k-space generative model(HKGM),which can generate samples from a training set of as little as one k-space.At the prior learning stage,we first construct a large Hankel matrix from k-space data,then extract multiple structured k-space patches from the Hankel matrix to capture the internal distribution among different patches.Extracting patches from a Hankel matrix enables the generative model to be learned from the redundant and low-rank data space.At the iterative reconstruction stage,the desired solution obeys the learned prior knowledge.The intermediate reconstruction solution is updated by taking it as the input of the generative model.The updated result is then alternatively operated by imposing low-rank penalty on its Hankel matrix and data consistency constraint on the measurement data.Experimental results confirmed that the internal statistics of patches within a single k-space data carry enough information for learning a powerful generative model and provide state-of-the-art reconstruction.The main contributions of this work are summarized as:(1)In the application of deep learning in magnetic resonance imaging research,a large number of data samples are often needed,and obtaining a large number of magnetic resonance data is very expensive and time-consuming.Moreover,the deep learning model is highly sensitive to both the quantity and quality of training data.In this paper,in order to alleviate the issue of insufficient data samples in prior learning,we first construct a Hankel matrix from the sampled data in k-space,then extract multiple k-space patches from the matrix.Therefore,a large number of training data can be obtained within one case of k-space data,so that we can overcome the problem that the training of a single data is very easy to lead to the occurrence of overfitting phenomenon and cannot conduct training,and learn the prior distribution of training data.(2)This paper proposes an adaptive iterative strategy.After the prior knowledge is learned,it is incorporated into the conditional generation for high quality reconstruction.In addition to sample generation,we impose low-rank penalty on the Hankel matrix and data consistency constraint on the measurement data alternatively at each iteration.Since the learned prior knowledge has no restriction on the channel number of reconstructed under-sampled data,and the trained model has strong generalization,it can be adopted to parallel magnetic resonance imaging reconstruction on different number of coils and different data.