Study of the suitability of polarimetric metrics to implement tissue classification models based on artificial intelligence

Abstract

Polarimetric images are used for the characterization of biological tissues as well as for the early detection of some diseases. Recently, it has been demonstrated that accurate classification models can be constructed based on polarimetric data, such as the Mueller matrix (MM) or different polarimetric metrics resulting from combinations of different MM elements. The choice of polarimetric observables to be used for classifying is usually arbitrary, but mathematical transformations from MM elements to other metrics may benefit or impair the accuracy of the final models. This work presents a thorough comparison of different classification models based on typical machine learning algorithms trained according to different polarimetric metrics, in the search of the most efficient polarimetric basis. The classification models are tested on different biological tissues obtained from a collection of ex-vivo chickens.