Cancer diagnosis system and method based on breast molybdenum target calcification characteristics

Abstract

The invention discloses a cancer diagnosis system and method based on breast molybdenum target calcification characteristics. The method comprises a first step of acquiring a breast molybdenum targetx-ray data set; a second step of carrying out image enhancement processing on each image in the breast molybdenum target x-ray data set respectively, and carrying out binaryzation segmentation to obtain a calcified lesion area; a third step of carrying out feature extraction on the calcified lesion area subjected to image enhancement processing and binaryzation respectively to obtain traditional features and depth features; a fourth step of performing typical correlation analysis on the traditional features and depth features, deleting the depth features with the low association weight with the traditional features, and saving the depth features which are closely related to the traditional features; a fifth steps of carrying out support vector machine linear classification model training for the saved depth features to obtain a classifier. According to the method, over-fitting of a traditional convolutional neural network on the calcification diagnosis of the breast molybdenum targetscan be effectively avoided, and automatic diagnosis of the molybdenum target image lesion at any resolution is realized.

Description

technical field [0001] The invention relates to the field of medical image processing, in particular to a cancer diagnosis system and method based on calcification characteristics of mammary gland mammography targets. Background technique [0002] Mammography is the preferred imaging method for the diagnosis of early breast cancer, and has become the most effective means of breast cancer screening. Mammography focuses on microcalcifications and masses, which are the two most important diagnostic criteria for breast cancer and have characteristic imaging findings. Although the widespread use of large-scale census and mammography has reduced the mortality rate of breast cancer by about 18-40%, there are still 15-20% missed diagnosis rates and higher false positive rates. The reasons are: 1. Mammography images lack contrast and layering, and it is difficult to fully display the characteristics of microcalcifications and masses; 2. Radiologists are easily affected by subjective...

AI Technical Summary

Problems solved by technology

[0015] Although the convolutional neural network has achieved good results on natural image sets, new problems have emerged in the field of medical images, because the data sets of medical images are far from the number of natural image data sets, and the depth Learning largely depends on a large data set so that the model can fit the data set we give it as much as possible. If the data set is too small, it will cause the problem of over-fitting in the deep model we get, that is, in the training The model gets the correct classification on the sample, but the performance on the test sample is not satisfactory

Images