Mueller matrix polarimetry for automated assessment of dewaxed pancreatic tissue using deep learning

Pancreatic cancer is a leading cause of cancer-related death with very few therapeutic and intervention options. In this context, Mueller matrix polarimetry (MMP) offers a promising method to better understand tissue composition. While MMP has shown potential in characterizing fresh tissue from various organs, the challenge of collecting and imaging surgically removed fresh tissue in a timely manner limits scalability. In contrast, formalin-fixed paraffin-embedded (FFPE) tissue is a routinely performed histopathology process and provides an extensive archival resource for subsequent tissue analysis studies. However, the impact of FFPE processing on MMP remains an open question. This study investigates whether MMP can reliably differentiate tissue composition after fixation, embedding, and deparaffinization, or <i>dewaxing</i>. To evaluate this, we collected 76 FFPE pancreatic tissue samples from 30 patients and applied a dewaxing protocol to expose tissue and restore optical access. MMP images were acquired from these dewaxed samples in reflection (16-degree phase angle) using a custom-built 16-wavelength (460 to 598 nm) imaging MMP. A convolutional neural network (CNN) was then trained to classify cancer and noncancer tissue from the MMP data, using histopathology annotations from a specialist gastrointestinal/visceral pathologist as ground truth validation. Our findings demonstrate that even after FFPE processing and dewaxing, sufficient polarization information is retained for effective tissue classification, with the CNN achieving an AUROC of 0.91 &plusmn; 0.08. This confirms that MMP is viable for the retrospective analysis of dewaxed FFPE archival samples. By validating this approach, this work establishes a practical, scalable foundation for leveraging existing pathology archives and supports the broader integration of polarization-based imaging into AI-driven histopathology and clinical diagnostics.