Physical realizability classification of incomplete Mueller matrices: toward real-time polarimetric imaging
Year: 2026
Type: article
Abstract: Translating wide-field imaging Mueller polarimetry to clinical settings requires video-rate data acquisition, achievable with polarization-sensitive division-of-focal-plane cameras. Such sensors measure only the first three rows of a sample’s Mueller matrix , yielding a partial 3 × 4 Mueller matrix. Physical realizability testing of Mueller matrix is a mandatory post-processing step, which analytical implementation requires the knowledge of complete 4 × 4 Mueller matrix elements. No closed-form physical realizability test exists for partial Mueller matrices. We present a supervised machine learning framework that tests physical realizability directly on partial Mueller matrices, without the missing fourth row. Three different classifiers – XGBoost, CatBoost, and Multi-Layer Perceptron – were trained and evaluated on experimental Mueller matrix images of brain, skin and cervical tissue and also on polarized Monte Carlo simulated Mueller matrix images of optical phantoms. On independent held-out complete MM images excluded from all stages of model development, the classifiers achieve mean image-level F 1 scores of 0.973–0.975 across the evaluated partial-MM inputs. We further show that reducing the input to the 3 × 3 top-left sub-matrix shows comparative performance, with all three classifiers generalizing comparably and the Multi-Layer Perceptron classifier offering the fastest inference.
Source: Biomedical Optics Express
DOI: 10.1364/boe.608210
Language: English
Subfield: Atomic and Molecular Physics, and Optics
Field: Physics and Astronomy
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $1,913
OpenAlex ID: W7213465166
