Abstract 2468: Automated co-registration of histology and mass spectrometry imaging for enhanced prostate cancer pathomics

Abstract Background: Prostate cancer (PCa) progression has been shown to be influenced by the relationship of extracellular matrix (ECM) proteins within the stroma, and immune and tumor cells. However, the current standard of care for PCa patients does not take these factors into account. Previously, our team identified sets of ECM glycans using matrix-assisted laser desorption/ionization (MALDI) imaging that were associated with PCa. We also found preliminary evidence that MALDI imaging can guide pathomic textural analysis in H&E images. However, our approach required manual registration between the MALDI and hematoxylin and eosin stain (H&E) images to derive pathomic textural features. In this study, we aim to fully automate the image co-registration and segmentation to guide the identification of pathomic textural features, and to demonstrate that glycan expression in MALDI images can identify regions of PCa in H&E images as tumor. Methods: We identified a cohort of PCa patients (n=10) who had undergone radical prostatectomy treatment (grade groups GG3-GG5), and with available H&E images, as well as five MALDI images per patient representing the distribution of N-glycan species at 1419, 2539, 2686, 1663, and 1809 m/z. The H&E and MALDI images were co-registered using feature-based deformable registration methods based on mutual information, and segmentations were automatically calculated using a computational imaging framework based on rolling ball, erosion, dilation, Otsu’s thresholding, and large connected component analysis. This created 6 regions of interest on the H&E image (5 N-glycans, 1 Pathologist) for analysis of pathomic textural features. Dice scores were used to measure the overlap between the glycan mask and the pathologist segmentation in H&E. We measured 70 pathomic textural features using PyRadiomics on the H&E images for the 5 glycan regions and compared every combination of textural features using Wilcoxon signed rank tests. Results: The median Dice scores between the N-glycan mask and pathology tumor segmentation were: 1419 (0.48 [0.39‒0.65]), 2539 (0.64 [0.44‒0.76]), 1663 (0.27 [0.15‒0.37]), 1809 (0.38 [0.14‒0.42]), and 2686 m/z (0.29 [0.09‒0.35]). This indicates the glycans with the largest overlap with tumors were 1419 and 2539 m/z, while glycans 1663, 2686 and 1809 m/z overlap most with non-tumor regions. Out of the 70 pathomic textural features, 55 were significantly different (p<0.05) between the masks for 1419 and 1663 m/z, 48 between 1419 and 1809 m/z, 47 between 2539 and 1663 m/z, and 50 between 2539 and 1809 m/z. In addition, there were only 10 significantly different features between 1419 and 2539 m/z, and 4 features between 1663 and 1809 m/z. Conclusions: We provided further evidence that tissue properties and pathomic textural analysis in H&E images can be guided by MALDI imaging, and preliminary evidence that the process can be fully automated. Citation Format: Jader Giraldo-Guzman, Cameron A. Czerpak, Cody Weimholdt, Lewis J. Thomas, Sheng-Kwei Song, Peggi M. Angel, Richard R. Drake, Joseph E. Ippolito, Jose Marcio Luna. Automated co-registration of histology and mass spectrometry imaging for enhanced prostate cancer pathomics [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 2468.