Abstract 2505: Semi-automated image registration and cell typing integrates multiplexed imaging data to investigate the tumor microenvironment in clinical biopsies

Abstract Introduction: Spatial profiling technologies are individually limited by the number of proteins evaluated, making it beneficial to integrate imaging analysis across multiple slides from a single biopsy. Here, we describe a semi-automated workflow for the scaled image registration and downstream analysis of clinical melanoma biopsies treated with immunotherapy and evaluated across multiplexed protein platforms. Methods: Clinical biopsies (n=119) from patients with melanoma (N=39) were assayed on the Vectra Polaris platform. Two sequential slides from each biopsy were individually processed under pre-specified custom named panels, being: PD1 panel (DAPI, CD8, PDL1, Ki67, PD1, SOX10, and CD68) or the L panel (DAPI, CTLA4, CD8, FOXP3, CD4, CD20, and PD1). The VALIS program was used to perform semi-automated image registration on these sequential slides. A bootstrapping approach determined cell marker positivity to identify cell types including melanoma cells (SOX10), CD8 T cells (CD8), CD4 T cells (CD4, FOXP3-), regulatory T cells (Tregs; CD4, FOXP3), B cells (CD20), and macrophages (CD68). Following image registration, cell type relationships were evaluated for regional annotation of tumor content, T-cell infiltration, and T-cell activation. Results: There were 9, 376, 260 cell segments detected across biopsies (average n=42, 235 segments per biopsy, range 1, 611-214, 050). Melanoma cells were detected based upon nuclear SOX10 staining. Biopsies had varied levels of tumor cellularity (average 8.9%, range 2.1-24.0%). Image registration was performed for samples that had corresponding PD1 and L panels (n=112). Automated registration was achieved for 86 of 112 paired samples (76.8%); 8 of the 86 (9.3%) required manual changes in parameters for accurate registration, and 18 samples were unable to be registered with VALIS due to imaging noise or poor-quality biopsies. This successful registration in the majority of cases led to the automated annotation across images, transferring of the PD1-panel image on to the registered L-panel image. Conclusions: Semi-automated image registration can help to overcome protein expression measurement limitations in spatial profiling technologies, enabling the expanded knowledge transfer across these data modalities. This workflow allows the effective integration of data measurements from multiple stainings for more comprehensive analysis using these technologies to study the tumor microenvironment. Citation Format: Cadence Chang, Egmidio Medina, Sarah Samordnitsky, Alan Zelin, Qin Zhou, Siwen Hu-LieskovanMichael C. Wu, Antoni Ribas, Katie M. Campbell. Semi-automated image registration and cell typing integrates multiplexed imaging data to investigate the tumor microenvironment in clinical biopsies [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 2505.