Symmetric residual U-shaped network breast mass segmentation method based on composite weighted loss function

Abstract

The invention discloses a breast lump segmentation method based on deep learning. According to the scheme, a deep neural network is trained by training a manually labeled lump image. After a complete mammary gland image is input, the network can autonomously learn imaging features of the lumps, and an output result is an area which is identified as the lumps by the network, so that end-to-end mammary gland lump segmentation is realized. In order to improve the detection rate of the lumps, the invention discloses a novel weighted compound loss function.

Description

technical field [0001] The invention relates to a deep learning image processing technology, and specifically designs a method for segmenting masses of mammography images. Background technique [0002] The mammary gland is composed of skin, fibrous tissue, mammary glands and fat. Breast cancer is a malignant tumor that occurs in glandular epithelial tissue. Among them, women accounted for 99%, and men accounted for only 1%. Early breast cancer usually has no obvious clinical symptoms, and mostly manifests as a painless mass or mild breast pain. The "China Breast Disease Survey Report" shows that only 5% of women undergo breast disease examination once a year, and 23% of women think that There is no need for special breast disease examination, coupled with the insufficient promotion of breast disease census in my country, most grassroots medical institutions lack professional medical personnel and census equipment, resulting in a low early detection rate of breast cancer in m...

AI Technical Summary

Problems solved by technology

However, the two-stage network relies too much on the accuracy of the detection process in terms of segmentation accuracy, and increases the complexity of the process, which is not in line with the stable and efficient development direction of computer-aided diagnosis systems.

Images