Fig. 1:

A schematic of the training of the convolutional neural network (CNN) to analyze brain proton MRS data. ‘Metabolite‐only’ spectra, simulated by linearly combining phantom spectra of individual compounds according to randomly selected relative concentration ratios, are used as the ground truth in the training of the CNN. To mimic in vivo brain spectra, the metabolite‐only spectra are combined with noise, line broadened, frequency and phase shifted, and a slowly varying baseline added. These simulated brain spectra are used as the input to the CNN in the training, whose goal is to determine the original metabolite-only spectra and the individual metabolite concentrations. Abbreviations: Conv, convolution; Batch Norm, batch normalization; Max Pool, max pooling; ReLU, rectified linear unit. Reproduced, with permission, from Lee H.H. and Kim H. [15].