Systems and methods for improving magnetic resonance imaging using deep learning

Abstract

A computer-implemented method is provided for improving image quality with shortened acquisition time. The method comprises: determining an accelerated image acquisition scheme for imaging a subject using a medical imaging apparatus; acquiring a medical image of the subject according to the accelerated image acquisition scheme using the medical imaging apparatus; applying a deep network model to the medical image to improve the quality of the medical image; and outputting an improved quality image of the subject, for analysis by a physician.

1 . (canceled) 
     
     
         2 . A computer-implemented method for improving image quality with shortened acquisition time, the method comprising:
 (a) receiving a medical image, wherein the medical image is acquired with a shortened acquisition time;   (b) applying a deep network model to the medical image to output a predicted medical image, wherein the predicted medical image has a quality higher than a quality of the medical image,   wherein the deep network model is trained using training datasets comprising at least a pair of high quality image and low quality image and by adaptively tuning one or more model parameters to estimate a function that transforms the low quality image to the high quality image.   
     
     
         3 . The computer-implemented method of  claim 2 , wherein the training datasets further comprise augmented datasets. 
     
     
         4 . The computer-implemented method of  claim 3 , wherein the augmented datasets comprise simulated low quality image. 
     
     
         5 . The computer-implemented method of  claim 4 , wherein the simulated low quality image is generated by applying a filter to or adding noise to a raw image data. 
     
     
         6 . The computer-implemented method of  claim 2 , wherein the training datasets comprise image patches that comprise a portion of at least the low quality image and the high quality image. 
     
     
         7 . The computer-implemented method of  claim 2 , wherein the function is a residual function. 
     
     
         8 . The computer-implemented method of  claim 2 , wherein the deep network model is trained using a residual learning framework. 
     
     
         9 . The computer-implemented method of  claim 2 , wherein the quality of the predicted medical image has greater signal-to-noise ratio (SNR), higher resolution, or less aliasing compared with the quality of the medical image. 
     
     
         10 . The computer-implemented method of  claim 2 , wherein the shortened acquisition time is achieved by undersampling k-space, reducing number of repetitions or reducing resolution. 
     
     
         11 . The computer-implemented method of  claim 2 , wherein the shortened acquisition time is at least 1.5 times, 2 times, 3 times, 4 times, 5 times faster than a standard acquisition time. 
     
     
         12 . A non-transitory computer-readable storage medium including instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising:
 (a) receiving a medical image, wherein the medical image is acquired with a shortened acquisition time;   (b) applying a deep network model to the medical image to output a predicted medical image, wherein the predicted medical image has a quality higher than a quality of the medical image,   wherein the deep network model is trained using training datasets comprising at least a pair of high quality image and low quality image and by adaptively tuning one or more model parameters to estimate a function that transforms the low quality image to the high quality image.   
     
     
         13 . The non-transitory computer-readable storage medium of  claim 12 , wherein the training datasets further comprise augmented datasets. 
     
     
         14 . The non-transitory computer-readable storage medium of  claim 13 , wherein the augmented datasets comprise simulated low quality image. 
     
     
         15 . The non-transitory computer-readable storage medium of  claim 14 , wherein the simulated low quality image is generated by applying a filter to or adding noise to a raw image data. 
     
     
         16 . The non-transitory computer-readable storage medium of  claim 12 , wherein the training datasets comprise image patches that comprise a portion of at least the low quality image and the high quality image. 
     
     
         17 . The non-transitory computer-readable storage medium of  claim 12 , wherein the function is a residual function. 
     
     
         18 . The non-transitory computer-readable storage medium of  claim 12 , wherein the deep network model is trained using a residual learning framework. 
     
     
         19 . The non-transitory computer-readable storage medium of  claim 12 , wherein the quality of the predicted medical image has greater signal-to-noise ratio (SNR), higher resolution, or less aliasing compared with the quality of the medical image. 
     
     
         20 . The non-transitory computer-readable storage medium of  claim 12 , wherein the shortened acquisition time is achieved by undersampling k-space, reducing number of repetitions or reducing resolution. 
     
     
         21 . The non-transitory computer-readable storage medium of  claim 12 , wherein the shortened acquisition time is at least 1.5 times, 2 times, 3 times, 4 times, 5 times faster than a standard acquisition time.