Organic tissue recognition through polarimetric-based algorithm

During the last decades, the attention on the application of polarimetric methods for biological tissues inspection has been increasing. Nowadays, organic tissue recognition algorithms are of potential interest in different research areas, as for instance, in biomedical applications for the early detection of diseases or the classification of biological structures. Based on the modifications in polarization that light-matter interactions produce, an exhaustive polarimetric analysis of the sample (extraction of dichroism, retardance and depolarization) may unveil the different tissue inherent characteristics and provide a complete description of how the biological structures interact with incident polarized light. By taking advantage of such polarimetric methods tissues characterization, we propose four predictive models corresponding to the recognition of four ex-vivo chicken tissue categories: bone, muscle, tendon and myotendinous junction tissue samples. The implemented multivariant probabilistic models are based on the logistic regression fit of the experimental Mueller matrix-derived polarimetric observables (measured at three different wavelengths: 625 nm, 530 nm and 470nm): polarizance P, diattenuation D, depolarization content (Indices of Polarimetric Purity P-1, P-2, P-3 and depolarization index P-Delta), retardance (global, R, and linear delta) and optical rotation Psi. As a result, we achieve stable predictive models whose output, in terms of sensitivity and specificity indicators, are of 82.6% and 80.6% for bone recognition, 85% and 93.5% for tendon, 86% and 88.8% for muscle and 82% and 71% for myotendinous junction, respectively. Obtained results suggest that these non-invasive methods could be applied in multiple biomedical scenarios such as for early diagnosis of pathologies.

Automated summary

In recent years, there has been an increasing attention on the application of polarimetric methods for biological tissues inspection. This study proposes four predictive models for the recognition of four ex-vivo chicken tissue categories. By analyzing the polarization characteristics, the models achieve stable and accurate results for the recognition of bone, tendon, muscle, and myotendinous junction tissues.