Breast image processing method based on sparse auto-encoding deep network

Abstract

The invention discloses a mammary gland image processing method based on a sparse automatic coding depth network. The method comprises the realizing steps: (1) inputting an image; (2) calculating a gray level co-occurrence matrix; (3) extracting the characteristics of the gray level co-occurrence matrix; (4) training the sparse automatic coding depth network; (5) training a classifier v-SVM (Support Vector Machine); (6) classifying. The characteristics of the mammary gland image provided by the invention are unlikely to be interfered by noise and are stronger in robustness, and the extracted characteristics are good in completeness and comprehensiveness after the completion of the training of the sparse automatic coding depth network, and the classification accuracy can be effectively improved in the course of classifying, therefore the method is a method of effectively identifying a suspicious pathologic region in the mammary gland image.

Description

technical field [0001] The invention belongs to the technical field of image processing, and further relates to a breast image processing method based on a sparse automatic coding deep network in the technical field of medical image processing. The present invention can be used to extract features of lesion areas in breast images such as X-ray images and CT images, and further classify and identify lesion areas and normal areas in the image, so as to realize the comparison between the lesion area and the normal area of ​​the image, and further provide Clinicians analyze the extent of organ lesions to provide image evidence. Background technique [0002] Breast cancer is one of the most common and high-incidence malignant tumors in women, seriously affecting women's physical and mental health and even threatening their lives. At present, the monitoring of breast cancer is increasingly dependent on computer-aided diagnosis technology, and the key to computer-aided diagnosis t...

AI Technical Summary

Problems solved by technology

Although this method is simple and easy to use, there are still shortcomings: this method only extracts the underlying feature representation of the lesion area in the breast image. Because the underlying feature structure is relatively simple and sensitive to noise and background, it cannot obtain a complete and comprehensive image. The feature representation of breast images leads to low classification accuracy of breast images and cannot effectively identify suspicious lesion areas in breast images

This method makes the lesion area in the image more prominent, but there are still shortcomings: in the feature representation of the visual attention model, only the grayscale feature of a single pixel in the lesion area in the breast image is used, so it cannot be fully represented The texture features of the mass area in the breast image, when classifying, some of the mass with thicker texture in the breast image may be missed

Images