Optimizing the classification of biological tissues using polarized data supported by Machine Learning

Polarimetric data is nowadays used in the biomedical field to inspect organic tissues or for the early detection of some pathologies. In this work, we present a thorough comparison between different classification models based on several sets of polarimetric data, this allowing us to choose the polarimetric framework to construct tissue classification models. Four different well-known machine learning models are compared by analyzing three polarimetric datasets: (i) a selection of ten representative polarimetric observables; (ii) the Mueller matrix elements; and (iii) the combination of (i) and (ii) datasets. The study is conducted on the experimental Mueller matrices images measured on different organic tissues: muscle, tendon, myotendinous junction and bone; all of them measured from a collection of 165 ex-vivo chicken thighs. Provided results show the potential of polarimetric datasets for classification of biological tissues and paves the way for future applications in biomedicine and clinical trials.