Knowledge graph construction method for breast cancer ultrasonic image high-confidence entity relationship

Abstract

The invention provides a knowledge graph construction method for a breast cancer ultrasonic image high-confidence entity relationship. The method comprises steps of: firstly, extracting entities of breast cancer ultrasonic images through a deep residual network; secondly, mining the sample-entity matrix by using a biclustering algorithm to obtain a plurality of highly associated sub-matrixes; secondly, constructing and solving an optimization function containing a confusion factor interference dependent variable relationship to obtain a high-confidence relationship set among entities; and finally, constructing a knowledge graph of the breast cancer ultrasonic image by taking the entities as nodes and taking the high-confidence relationship between the entities as an adjacent edge. According to the method, the high-confidence entity is mined from the breast cancer ultrasonic image and constructed into the knowledge graph, a basis can be provided for interpretable diagnostic reasoning, the problem that the confidence degree of a prediction result is not high due to few medical data samples can be well solved, and the method is more suitable for mining a small-sample high-confidence entity relationship of medical ultrasonic data.

Description

technical field [0001] The invention belongs to the field of ultrasound image analysis of breast cancer, and in particular relates to a method for constructing a knowledge map of high-confidence entity relationships in ultrasound images of breast cancer. Background technique [0002] Medical ultrasound is the most commonly used imaging method in clinical practice. When dealing with large-scale breast screening, the number of experienced ultrasound doctors is obviously insufficient. In order to reduce the dependence of large-scale screening on a large number of doctors, computer-aided diagnosis is currently mainly introduced into medical ultrasound, and mining relevant knowledge from ultrasound images is an important prerequisite for realizing computer-aided diagnosis. [0003] In the existing research on breast ultrasound knowledge mining, it mainly focuses on segmenting the tissue structure area in the ultrasound image. Among them, Xian et al. in the literature "Xian M, Zh...

AI Technical Summary

Problems solved by technology

The invention digs out high-confidence entities from breast cancer ultrasound images and builds them into knowledge graphs, which can provide a basis for interpretable diagnostic reasoning, and can better solve the problem of low confidence in prediction results caused by fewer medical data samples , which is more suitable for the mining of small-sample high-confidence entity relationships in medical ultrasound data

Images