Cancer Tissue Recognition by Muller Matrix Imaging Full Array Curve

As the second most common cause of death in the world, cancer has become a persistent public health challenge facing the world. At present, the most mature treatment for cancer is surgical resection, and the determination of whether the resection is clean depending on the detection of artificial pathological sections. During the operation, it is necessary to repeatedly confirm whether the diseased tissue is removed cleanly, and each test requires the preparation of pathological sections for manual detection. Pathological section staining is tedious and time-consuming, and the process of manual section detection and confirmation is time-consuming and inefficient. Therefore, there is an urgent need for an auxiliary detection method that skips the staining process, and the method is accurate, fast, highly sensitive, non-contact, non-invasive, and highly safe. Polarization is a new information dimension independent of light intensity, wavelength, and phase. Polarization imaging uses the incident light to irradiate the sample to detect the outgoing light and obtain the physical information of the sample. Polarization imaging is a non-invasive, label free, non-contact optical detection technology. The detected polarization information is often represented by Mueller matrix. Mueller matrix characterizes the polarization characteristics of complex media, including rich macro and micro structural information. Mueller matrix imaging can encode rich information on microstructure even under low resolution and wide field conditions. Polarization detection technology has the advantages such as large information, strong compatibility, wide scale, but the microstructure of each element value of Mueller matrix is not directly related to the same product, and it is difficult to characterize due to the mixing of a large number of influencing factors. It is difficult to obtain specific indicators of sample microstructure. Today, the focus on information extraction of Mueller matrix is only on the processing level of individual array elements with obviously different information, which is very dependent on the information content of sample array elements. This research mainly uses the existing clear classification of physical relations to establish the relationship between the internal information of Mueller matrix, analyze the full array element of Mueller matrix vertically with pixel as the unit, and further extract the overall difference information. In this paper, we will start with the reconstruction of the original backward Mueller matrix imaging system light source and the incident direction to improve the imaging quality. We use the measurement method of double wave plates rotating 30 times to generate Mueller matrix, which is different from the previous data analysis and processing methods. With the help of the classified physical information of Mueller matrix, the new arrangement order is linked, and the Mueller matrix data cube is generated by analogy with the spectral data processing method in the remote sensing field. The full array element curve is extracted longitudinally in pixels. The whole array element curve calls the overall polarization characteristics, which is different from the description form of two-dimensional space image. The expression form is flexible and changeable, which can more intuitively describe the multi-level analysis ability of information. The included angle cosine method is used to eliminate the system error caused by the included angle of the light path, and the target classification is realized by traversing the entire two-dimensional space. This method is fast and easy to implement. On this basis, the samples were selected from the pathological sections of unstained lung cancer and breast cancer. According to the above methods, information visualization analysis, feature extraction and classification, cross scale experiments were carried out, and a confusion matrix was used to show the classification and accuracy. The results show that the pixel level full array information observation mode has more flexible target selection, more variable observation forms and strong anti-interference ability, and good accuracy in distinguishing cancerous change, non cancerous change and background. The accuracy of lung cancer and breast cancer can reach 89.594% and 87.82%, and the accuracy of high imagnification breast cancer is 77.52%. Low imagnification imaging with cross scale characteristics can still complete feature extraction. Skipping the staining process and assisting in manual detection is of great significance for shortening the time of pathological detection in clinical practice.