Machine Learning Approach to 3×4 Mueller Polarimetry for Complete Reconstruction of Diagnostic Polarimetric Images of Biological Tissues
Conference
Chae, S, Huang, T, Rodríguez-Núñez, O et al. (2025). Machine Learning Approach to 3×4 Mueller Polarimetry for Complete Reconstruction of Diagnostic Polarimetric Images of Biological Tissues
. 13322 10.1117/12.3042830
Chae, S, Huang, T, Rodríguez-Núñez, O et al. (2025). Machine Learning Approach to 3×4 Mueller Polarimetry for Complete Reconstruction of Diagnostic Polarimetric Images of Biological Tissues
. 13322 10.1117/12.3042830
We present a machine learning (ML) algorithm to reconstruct the missing row of partial 3×4 Mueller matrix (MM) images of biological tissues. Using a sequential neural network model, we trained and tested our algorithm on large polarimetric datasets of the complete 4×4 MM images acquired on the excised human tissues (cervix, colon, skin, brain) with two different imaging Mueller polarimeters operating in reflection and transmission configurations, at different wavelengths (550 nm and 385 nm, respectively). Physical realizability filtering of MM data was applied prior to training to remove erroneous or noisy pixels. The results demonstrate an accurate reconstruction of the missing MM elements. We further confirmed that the polarimetric maps of diattenuation, retardance and depolarization, obtained from the Lu-Chipman decomposition of reconstructed MMs, are nearly identical to those obtained from the original complete MMs. Our findings suggest that combining the fast acquisition of partial MM with polarization sensitive camera and deep learning-based reconstruction model could enable real-time diagnostic polarimetric imaging in clinical settings.
