The Reconstruction Method Using Compressed Sensing and Convolutional Neural Network for PROPELLER MRI in Head

PROPELLER MRI is a method of reconstruction from the collected data of rectangular regions (blades) rotating around the origin of the k-space. This method can compensate for the motion of the subject by using the phase and rotation between the blades. Collecting more blades will improve the accuracy of the correction but will increase the imaging time. On the other hand, reducing the number of phase encodings and blades for shortening the imaging time results in streak artifacts. For this study, we attempted to improve the image quality of the PROPELLER MRI in head by reconstructing using less data with compressed sensing (CS) and a convolutional neural network (CNN), which is a deep-learning method. For k-space data with sampling rates of 11% to 54% by varying the width and number of blades, we compared three patterns of the reconstruction method: A) with only CS, B) with only CNN, and C) with both CS and CNN. For all sampling rates, the method with CS and CNN yielded the best evaluation value; therefore, it is suggested that the image quality could be improved by reconstructing using CS and CNN when the sampling rate is low.