Single-sequence MRI Glioma Grading Based on Semi-supervised Learning and Decision Vector Alignment
A method based on semi-supervised learning and decision vector alignment was proposed for glioma grading in single-sequence magnetic resonance imaging(MRI).In this method,the cosine similarity loss function was used to guide the proposed convolutional neural network model to complete the missing sequence features through the single-sequence MRI data,and the classification decision vector aligned with the multi-sequence MRI was obtained.Considering the scarcity of high-quality graded labeled data,a semi-supervised learning strategy was introduced to make full use of labelled and unlabelled MRI data.This strategy could simultaneously train and optimize the alignment of decision vectors and the single-sequence MRI glioma grading model,to improve the grading effect of the model.The performance of the proposed method was evaluated by using five-fold cross validation.The experimental results on the BraTS dataset showed that the proposed method achieved 91.64％accuracy in the glioma grading task based on T1ce sequence,which was superior to the existing methods.
万方数据
维普
