Detection of bone and soft tissue tumors using Mueller matrix polarimetry: a non-invasive approach for unfixed tumor identification

Mueller matrix polarimetry (MMP) effectively captures the polarization characteristics of biological tissues. This study established a subcutaneous osteosarcoma model in mice, utilizing MMP to analyze fresh tumor samples. We selected multilayer perceptron (MLP) and Kolmogorov-Arnold networks (KAN) as our classifiers and compared them based on nearby pixel attributes and single-pixel attributes, using MMP parameters as inputs. KAN generally performs better than MLP, suggesting its great potential for MMP datasets. Mueller matrix’s diagonal elements and the parameters Δ and D significantly contributed to tumor diagnosis. This research explores a non-invasive method for unfixed tumor identification, aiming for intraoperative rapid diagnosis.