Medical key point detection method and device based on multi-scale topological graph and application

Abstract

The invention provides a medical key point detection method and device based on a multi-scale topological graph and an application, wherein the method comprises the steps: extracting a self-vector of a key point based on a thermodynamic diagram of a key point; constructing a multi-scale adjacent matrix based on the self-vector; distributing different weights for different matrixes in the multi-scale adjacent matrix, carrying out training, and thus obtaining a key point matrix; and obtaining and determining the key points based on the key point matrix. Therefore, topological points consider the positions of other topological points when paying attention to the positions of the topological points, the structure of a topological graph is continuously adjusted, and finally the effect that the topological graph can be qualified for a key point detection task is achieved so as to improve the accuracy of key point detection.

Description

technical field [0001] The invention relates to the field of image key point detection, in particular to a medical key point detection method, device and application based on multi-scale topological graphs. Background technique [0002] Medical image analysis is a cross-analysis method that integrates technologies such as medical imaging, mathematical modeling, mathematical image processing and analysis, artificial intelligence, and numerical algorithms. It provides an auxiliary reference for the clinical diagnosis and treatment of various diseases. contribute to the advancement of clinical medicine. [0003] Accurate positioning of medical key points has always been the focus and difficulty of medical image analysis. The accuracy of key point positioning directly affects the results of medical image analysis. For example, in the analysis of spine images, it is necessary to accurately obtain the position of vertebral body key points. , in the analysis of thoracic images, it...

AI Technical Summary

Problems solved by technology

The traditional way of subjectively judging the key points of medical images has many problems such as time-consuming, labor-intensive, low accuracy, etc., and it is difficult to meet the requirements of high-precision medical image analysis.

In recent years, artificial intelligence image analysis models based on deep convolutional neural networks have been able to detect key points of medical images through heat map regression. The lack of information between keypoints and keypoints cannot represent the global anatomical structure and constrain the relational structure between keypoints, resulting in the inability of convolutional neural networks to provide high-level structural information for individual keypoints

[0005] See CN112541893A "A method for detecting bifurcation relationship points in a tree-like structure in a three-dimensional tomographic image" applied by Tsinghua University, although it also considers the relative positional relationship between the key points when detecting medical key points, and it solves the problem The method is to enable the fully convolutional deep learning network to simultaneously complete the three subtasks of bifurcation key point detection, tree structure segmentation, and branch vector field regression. It is suitable for the detection of bifurcation key points in the anatomical tree structure, but not suitable for the detection of other medical key points with complex and disordered key point relationships. This document focuses more on the key point detection of tree structure bifurcations

[0006] It can be seen that in recent years, major universities and research institutions have been working on the accurate detection of key points. Although there are also research reports that introduce the adjacent relationship of key points into the detection of key points, the applicable scenarios of detection are limited. And the problem of lack of accuracy of key points

Images