Combination of Muller matrix imaging polarimetry and artificial intelligence for classification of mice skin cancer tissue in-vitro and in-vivo

Abstract

Mueller matrix imaging polarimetry is a fast and non-invasive technique for discriminating between different types of biological samples based on the characteristics of polarized light interacting with them. Combining Mueller matrix imaging polarimetry with artificial intelligence provides further advantages in detecting different kinds of medical conditions in an automated manner. Accordingly, the present study proposes a method based on Mueller matrix polarimetry and machine learning algorithms for discriminating between (1) four different types of mice skin tissues (normal, acanthosis, papilloma, and squamous cell carcinoma); (2) two types of mice skin tissues which show histological similarities to human equivalents (normal and squamous cell carcinoma); and (3) 7,12-dimethylbenz[a]anthracene/estrogen-induced mice skin tissues. Five machine learning classifiers, namely Random Forest, Support Vector Machine, K-Nearest Neighbor, Gradient Boosting, and Gaussian Naïve Bayes, are considered in the first and second applications, while three models (Support Vector Machine, K-Nearest Neighbor, Gradient Boosting) are considered in the third application. For each application, the features which dominate the machine learning prediction performance are determined through multivariate correlation matrix analysis, kernel density estimation, and Analysis of Variance tests. The experimental results show that the Random Forest model achieves the highest classification accuracy (93.55 %) for the first application, while the Support Vector Machine model yields the highest accuracy for both the second and third applications (97.66 % and 100 %, respectively). Overall, the proposed framework consisting of Mueller matrix imaging polarimetry and machine learning provides a strong foundation for the on-going development of screening and diagnosis methods for human skin cancer.