Quantitative Analysis Of Breast Cancer Tissue Based On Mueller Matrix Images

Breast cancer is one of the malignant tumors that pose a serious threat to women’s health.Rapid and accurate diagnosis is an effective way to reduce breast cancer mortality.Compared with other cancers,breast cancer is highly dependent on rapid intraoperative pathology.Mueller matrix parameters have shown good diagnostic potential in characterizing the microstructure and optical properties of various cancer pathological tissues,such as skin cancer,breast cancer,cervical cancer,etc.Digital image analysis methods are widely used in the field of cancer pathology diagnosis due to the advantages of objectivity and accuracy.Based on the breast cancer images obtained by the Mueller matrix imaging system,the segmentation and quantitative analysis of cancerous tissue and normal tissue is one of the important steps in the pathological diagnosis of breast cancer.Therefore,to provide a repeatable,objective and reliable quantitative analysis method based on Mueller matrix images has important research significance for the rapid pathological diagnosis of breast cancer and other cancers.In this paper,a set of backscattered light Mueller matrix automatic imaging system is used to image unstained pathological tissue sections,and the corresponding parameter images are obtained by using Mueller matrix decomposition and transformation.Experiments and analysis are carried out on the image smoothing,weighted average fusion,data expansion and other preprocessing methods of parametric images.Since the pathological sections of breast cancer are rare and precious,in order to improve the universality and effectiveness of the method in this paper,the research on the pathological tissue of liver cancer and lung cancer is also increased.Aiming at the problem of time-consuming and lack of consistency in manual labeling and quantitative analysis of cancerous and normal tissues in medical images,a semi-supervised semantic segmentation network model based on generative adversarial network was proposed,in which the improved Seg Net semantic segmentation network was used as the generator,using the residual network module to learn the parameter images to enhance the image details,and obtain good segmentation results of cancerous and normal tissues.The results of segmentation experiments showed that,compared with traditional segmentation algorithms,this algorithm achieved good results in both subjective and objective evaluation indicators,can automatically learn the characteristics of cancerous and normal tissues,and reduces the false negative rate of cancerous tissues.Based on the different regions obtained from the segmentation results,the feature extraction of gray level co-occurrence matrix,Tamura texture parameters and central moment parameters was performed,and the difference between cancerous and normal tissues after segmentation was quantitatively analyzed by combining with the content of lesion components.The analysis results showed that the standard deviation,contrast and directionality of the Mueller matrix transformation parameter image t₂ of breast cancer and lung cancer,and the skewness,roughness and contrast of the Mueller matrix transformation parameter image ₃t of liver cancer were significantly different between cancerous and normal tissue regions.The difference and the amount of lesions in the pathological tissue can further provide a theoretical basis for clinicians to diagnose cancer in real time and accurately pathologically.