Digital pathology uses digitized pathological images and their features in conjunction with artificial intelligence technology to achieve quantitative characterization of cancerous tissues and assist pathologists in clinical diagnoses. The use of polarized light illumination and polarized light detection can achieve full polarization imaging. Accordingly, the polarization characteristics of each pixel of the image contain abundant microstructural information, especially subcellular super-resolution information, that is difficult to obtain with nonpolarization imaging. Polarization imaging can provide a more effective means for the identification and quantitative characterization of cancerous tissues. This paper introduces Mueller matrix microscopic imaging techniques and comprehensively reviews the latest methods for polarization feature extraction, including supervised learning-based polarization pixel and image feature extraction, unsupervised learning-based polarization pixel clustering, and the extension of annotations through polarization feature templates based on super-pixels, highlighting their potential clinical applications.

Mueller matrix imaging provides abundant subcellular-level information on tissue microstructures. The quantitative extraction of polarization features from Mueller pixels is crucial for the clinical application of polarization imaging. In contrast to stain image-based digital pathology, polarization feature extraction through supervised learning offers more abundant microstructural information. However, the reliance on extensive, well-annotated data poses time and labor challenges. Moreover, supervised learning is dependent on pathologists’ prior knowledge, limiting the comprehensive utilization of information from the polarization space. Unsupervised clustering methods facilitate the decomposition of pathological tissues into distinct microstructural subtypes, enhancing the exploration of the rich information embedded in Mueller pixels. Additionally, this approach provides evidence for the ongoing discovery of new physical properties, structural characteristics, and dynamic processes at all levels above the subcellular scale in organisms, including living entities.

Following advancements in molecular biology techniques, the specific identification of molecular components in biological entities is becoming a pivotal tool in biomedical research, thus leading to diverse omics approaches. Polarization-based digital pathology can leverage feature extraction methods developed in various omics approaches. The unsupervised clustering of Mueller pixels quantitatively extracts information at various levels above the subwavelength scale, enabling the integration of label-free, noninvasive, abundant information features of Mueller matrix imaging into novel spatiotemporal omics methods.