Dual-encoder semi-supervised cardiac mri biventricular segmentation method based on improved sam

Abstract

The application belongs to the technical field of medical image processing and specifically relates to a double-encoder semi-supervised biventricular segmentation method for cardiac MRI based on improved SAM. The teacher model and the student model used both comprise a double encoder, a prompt encoder and a mask decoder, the double encoder comprises a parallel ViT branch and a CNN-Mamba branch, the ViT branch realizes parameter fine-tuning through a position adapter and a feature adapter, the CNN-Mamba branch mainly comprises a convolution fusion Mamba block and a second-order dimension attention block, the convolution fusion Mamba block introduces an enhanced convolution block on the basis of a VSS block, and the second-order dimension attention block utilizes a covariance matrix to mine second-order statistical information of data; the output features of the ViT branch and the CNN-Mamba branch are weighted and summed to obtain output features of the double encoder; the output features of the double encoder and the prompt encoder are input into the mask decoder for decoding to obtain a segmentation result; the teacher model and the student model are jointly trained based on interpolation consistency regularization semi-supervision, the trained student model is used as a segmentation model and is used for biventricular segmentation of cardiac MRI. The segmentation model has strong feature extraction capability, can effectively extract edge information of a ventricle and boundary information of a biventricular connection area, and improves segmentation precision.

Images