Breast cancer lesion image segmentation method based on an end-to-end neural network

Abstract

The invention discloses a breast cancer lesion image segmentation method based on an end-to-end neural network. The breast cancer lesion image segmentation method comprises the following steps: step 1, labeling by hands: marking a lesion area on data according to experience of a doctor; step 2, preprocessing data: preprocessing the three-dimensional image data, and cutting the three-dimensional image data into an image size suitable for a neural network; step 3, selecting a model: selecting an improved end-to-end neural network model; step 4, training a model: training an optimal model on themarked training set by using an improved network model; step 5, predicting data: predicting a focus area by using the trained model; and step 6, evaluating a result: measuring the accuracy of focus segmentation by utilizing corresponding evaluation indexes. According to the improved end-to-end neural network provided by the invention, a hole residual network is used on an original resolution image, and rich feature information of a target is maintained; and a Dice loss function in a weighted form is also adopted, so that the segmentation effect on a small target and a lesion edge region is better.

Description

technical field [0001] The invention specifically relates to a method for segmenting images of breast cancer lesions based on an improved end-to-end neural network model. Background technique [0002] According to the relevant data of the World Health Organization, breast cancer has become the cancer with the highest incidence rate in women, seriously endangering women's health. In order to be able to reduce the mortality rate of breast cancer, early diagnosis and treatment are needed. [0003] With the continuous development of neural networks, three-dimensional convolutional neural networks have begun to emerge, which make full use of the inter-layer correlation of three-dimensional images, and have been better applied in the processing of 3D medical images. With the continuous development of deep learning, there are already some segmentation algorithm models for processing 3D images, such as 3D-Unet and V-Net. The two 3D segmentation algorithm models contain the prototy...

AI Technical Summary

Problems solved by technology

From the experimental comparison results, it is found that these algorithms still have some problems in the segmentation of the edge of breast cancer lesions. The boundary of edge segmentation is not clear enough, the segmentation effect of the edge soft tissue with similar gray level is not good, and the edge position is not accurate enough.

Images