A method for automatic steering and segmentation of left ventricle in pet three-dimensional image

Abstract

The invention discloses a method for automatic steering and segmentation of the left ventricle of a PET three-dimensional image. By constructing and training an encoder composed of a convolution module, a plurality of residual-convolution modules and a downsampling module, a space transformation network, including Multiple up-sampling modules and residual-convolution module decoders and PET 3D image automatic steering and segmentation models of the left ventricle with skip connections between encoders and decoders, using conventional view A as model input, using this The model automatically rotates the image to the clinical standard view and obtains the segmentation result of the left ventricular structure in this view based on the standard view. The present invention uses the deep learning network of multi-task learning to extract the position features and semantic features of the image, realizes the automatic steering from different angles to the standard view, heart positioning and structural segmentation of the left ventricle, and the one-stop processing operation reduces manual steering and segmentation The complexity and human error are improved, and the convenience and accuracy of image operation are improved.

Description

technical field [0001] The present invention relates to the fields of medical imaging and deep learning, in particular to a method for automatic steering and segmentation of the left ventricle of a PET three-dimensional image based on a deep learning network. Background technique [0002] PET cardiac imaging is currently an effective means for clinical diagnosis of myocardial viability and other cardiovascular diseases, efficacy evaluation, and prognosis judgment. It can non-invasively provide functional activity information of myocardial tissue to detect potential lesions that have not yet caused structural changes. Early diagnosis of vascular diseases is of great significance. During clinical PET heart examination, since the long axis of the heart is not parallel to the long axis of the human body, the doctor needs to perform relevant operations to rotate the heart in the image to the standard short-axis view for clinical diagnosis, and perform subsequent correlation analy...

AI Technical Summary

Problems solved by technology

[0003] The above related operations usually need to be manually operated by doctors. Such subjective operations have poor repeatability and high errors, which have a great impact on the accuracy of subsequent analysis.

In order to solve these problems, researchers have proposed related automatic steering and segmentation operations. However, the related algorithms have high requirements on the structural integrity of the left ventricle of the heart, and are prone to failure in patients with severe myocardial ischemia.

Based on deep learning technology, some scholars have also proposed a model for automatic steering and segmentation of SPECT cardiac images using deep learning. However, due to the large size of the three-dimensional image, the method requires too high hardware requirements for the computer system during operation, and it is difficult to meet the daily needs. computer processing needs

Images