Deep learning neural network cascade model-based human heart coronary artery extraction method

Abstract

The invention discloses a deep learning neural network cascade model-based human heart coronary artery extraction method. The method comprises the following steps of: S1, converting a CT sequence original picture into a picture format according to a certain window width and a certain window level so as to obtain a CT sequence picture; S2, segmenting the CT sequence picture through a pre-trained full figure model so as to obtain segmentation results of main coronary artery and main branch blood vessels; S3, extracting foreground pixels, at the current layer, of the blood vessels on the basis ofthe full figure-based segmentation results in step S2, calculating the center of each blood vessel at the current layer, extending a patch image according to positions, of a picture in an adjacent layer, corresponding to the center positions of the blood vessels, and segmenting the patch image through a pre-trained local patch model so as to obtain segmentation results of small blood vessels; andS4, fusing the segmentation results of the main coronary artery, the branch blood vessels and the small blood vessels so as to obtain a human heart coronary artery. The method has obvious advantagesfor the effect of segmenting the small vessels, and is capable of fusing the extracted small blood vessels with the main coronary artery and the main branch blood vessels so as to obtain a complete and clear heart coronary artery extraction result.

Description

technical field [0001] The present invention relates to image segmentation, in particular to a human heart coronary artery extraction method based on a deep learning neural network cascade model. Background technique [0002] Extracting coronary arteries from CT image sequences has important clinical value and practical significance. Affected by image quality, case variability, few effective pixels in small blood vessels, and interference from other tissue structures, it is a great challenge to achieve accurate extraction of coronary arteries. Traditional extraction methods are mainly based on enhanced filtering and region growing methods. Due to the influence of complex threshold parameter adjustments, the traditional methods have poor adaptability and anti-interference ability to different cases, and there are obvious small blood vessel omissions, veins or other tissues. The problem of being mis-segmented into coronary arteries. [0003] With the increasingly extensive r...

AI Technical Summary

Problems solved by technology

[0004] However, in existing coronary artery segmentation methods, only one full-image segmentation model is used for coronary artery segmentation, such as figure 1 The proportion of small blood vessels in the CT image is shown as an example. Since the small blood vessels (the area in the rectangular frame) are small targets with low contrast in the full image field of view, in the existing segmentation method based on deep learning, after two Down After sampling, the small blood vessels are basically too small to be recognized, and the segmentation results often show the absence of small blood vessels.

Images