The Research On Segmentation And Classification Technology Of Breast Infrared Thermal Image

As one of the most common malignant tumors,breast cancer has been threatening human life and health worldwide.Studies have shown that early diagnosis and treatment can reduce the harm caused by the disease and reduce mortality.At present,the existing methods for screening breast cancer in medicine are mainly X-ray,CT,etc.,but for young women,due to the low contrast caused by dense breasts,X-ray and other methods are not effective,and are easy to cause radioactive damage to the body.Infrared thermography detection has the advantages of non-invasive and non-contact.It can check and identify breast cancer according to the thermal characteristics of abnormal breast sites,and plays an important role in early clinical diagnosis.Therefore,this paper conducts in-depth research on breast infrared thermography,realizes automatic asymmetric region segmentation of regions of interest,obtains left and right breast images,and then uses computer image classification algorithm to identify abnormal breasts,which provides a basis for the analysis and detection of breast cancer in clinical practice.The main research contents are as follows:1)Breast infrared thermography data preprocessing and region segmentation.Firstly,the non-local mean filtering algorithm is used to denoise the grayscale image.Secondly,an adaptive shear function optimization histogram equalization algorithm is proposed to realize image enhancement.Finally,a fully automatic asymmetric region of interest segmentation method is proposed.The edge detection algorithm is optimized by morphological closure technology to detect the edge contour of breast heat map.Boundary segmentation based on contour projection;combined with boundary tracking and curve fitting technology,the left and right milk bifurcation points are found,and the left and right milk regions are segmented based on the bifurcation points.2)According to the feature distribution of breast infrared thermography,11 features such as entropy,correlation,variance,short run factor and long run factor are extracted and analyzed by using gray level co-occurrence matrix and gray level run matrix.The classification model of breast infrared thermography based on support vector machine is designed,and the performance of the trained classifier is verified.Finally,the accuracy of91% is achieved.This method is higher than the results of other similar classification methods,and realizes the classification function of normal and abnormal breast infrared thermography based on support vector machine.3)A multi-view breast infrared thermography classification model MVR-Net based on ResNet-18/34 network is proposed.Specifically,through the four-channel parallel network,the breast infrared thermal images on both sides of the front and 45 ° direction obtained by region segmentation are processed at the same time.Finally,the feature extraction and classification decision are integrated through the fusion layer to achieve breast anomaly classification.The accuracy of MVR-Net-18/34 proposed in this paper is 95.4% and 97.6%respectively,and the effectiveness of MVR-Net classification model is verified by a large number of experiments.