Method and device for identifying lumen contour and external elastic membrane of intravascular ultrasonic image and computer storage medium

Abstract

The invention discloses a lumen contour and external elastic membrane identification method and device for an intravascular ultrasonic image and a computer storage medium, the overall technical scheme is based on a contour segmentation network of a multi-task learning mechanism, a contour identification task is introduced on the premise of not adding training samples or other labeling work, and the accuracy of the contour identification is improved. The contour of the training sample can be obtained according to a threshold value method, the contour recognition precision can be improved by introducing contour constraint, contour information can be extracted by a network, the contour segmentation result is optimized, the precision of a network model is improved by introducing a multi-task learning mechanism, and the method is suitable for large-scale popularization and application. A multi-task learning mechanism of contour structure information is introduced by accurately identifying the distance of the contour, and the plaque load is calculated in combination with the contour of the lumen and the outer elastic membrane, so that the severity of coronary artery stenosis is observed, the plaque size is judged, and an operation scheme is guided and optimized.

Description

technical field [0001] The invention belongs to the field of medical and health technologies, and in particular relates to a method, device and storage medium for automatic identification of lumen outline and external elastic membrane based on intravascular ultrasonic images. Background technique [0002] Intravascular Ultrasound (IVUS), an imaging method of medical imaging, is mainly used in the diagnosis of coronary heart disease in recent years, and has achieved relatively good results in the field of clinical application. IVUS is an imaging technology based on ultrasonic echo detection. It uses the echogenic properties of the inner wall of the blood vessel, the atherosclerosis tissue in the wall, and the connective tissue covering the surface of the blood vessel, so that the output sound waves can return to the ultrasonic detection display and be visible. On the other hand, due to the sound permeability of the blood itself and the healthy muscle tissue of the blood vesse...

AI Technical Summary

Problems solved by technology

[0006] In order to overcome the defect of incomplete structural information caused by unclear identification of edge information, and to solve the problems of robustness, stability, and low accuracy, the technical solution proposed by the present invention is to construct a combined lumen and external elastic force based on IVUS images. The film recognition framework uses a multi-task learning mechanism to learn multiple related tasks in parallel through multi-layer convolution and pooling, output mask and contour information, and improve the generalization ability of the network; and use the shape structure prior of the target , which alleviates the loss of structured information caused by downsampling when the network is learning, and reduces its outliers. Because the proportions of samples of different categories are very different, it causes a major interference problem to the learning process of the algorithm, resulting in the defect of class imbalance. , using the shape prior to make up for the impact of the difference in the number of training samples of different categories in the classification task on the accuracy, that is, to alleviate the problem of category imbalance and improve the accuracy of multi-instance target segmentation; in addition, the newly added shape structure information extraction task The learning label does not require additional annotations and can be obtained directly from the segmentation labels of the training samples; moreover, the multi-task recognition model proposed in this application can be added to any segmentation network to achieve the effect of improving network performance with only a small number of parameters

Images