DEEP LEARNING-BASED MAGNETIC RESONANCE SPECTROSCOPY RECONSTRUCTION METHOD

摘要

Provided is a new method for reconstructing a complete spectroscopy from undersampled magnetic resonance spectroscopy data by using a deep learning network. First, a time signal is generated by using a finite exponential function, an aliased spectroscopy of a frequency domain is obtained after an undersampling operation is completed in the time domain, and the aliased spectroscopy and a complete spectroscopy corresponding to full sampling together form a training data set. Then, a data verification convolutional neural network model used for magnetic resonance spectroscopy reconstruction is established, and neural network parameters are trained by using the training data set to form a trained neural network. Finally, an undersampling magnetic resonance spectroscopy signal which needs to be reconstructed is inputted into to the trained data verification convolutional neural network to reconstruct a complete magnetic resonance spectroscopy. The present method for reconstructing the magnetic resonance spectroscopy by beans of a data verification convolutional neural network features a fast reconstruction speed and high reconstructed spectroscopy quality.