FULLY CONVOLUTIONAL NETWORK MODEL TRAINING METHOD FOR SPLITTING ABNORMAL SIGNAL REGION IN MRI IMAGE

摘要

A fully convolutional network model training method for splitting an abnormal signal region in an MRI image. The method comprises: obtaining an MRI sample image, and obtaining an abnormal signal region split sample image obtained after an abnormal signal region split is performed on the MRI sample image (S101); initializing a weight parameter of a fully convolutional network model (S102); and training the fully convolutional network model by using the MRI sample image and the split sample image as training samples, so as to obtain the fully convolutional network model used for splitting an abnormal signal region in an MRI image (S103). The method can resolve the problem in the prior art of wrong splits due to very tedious and time consuming manual split marking that is easily affected by subjective factors, and a precise split result can be efficiently generated without any image preprocessing and postprocessing steps.