Design and Implementation of a Spatial Biology Image-Analysis Tool for Multiplex Microscopy

Identifier:  TFM:2414

Handle: https://hdl.handle.net/20.500.11797/TFM2414

Authors:  Anchuela Cantarero, Daniel

Abstract:
This work presents an integrated and validated framework for spatial tumor analysis that combines two complementary approaches. First, a modular Python pipeline is developed to process multiplex fluorescence images with an arbitrary number of channels, generated by cyclic immunofluorescence platforms. The automated workflow includes instance cell segmentation (Cellpose), cytoplasm estimation, adaptive intensity quantification, single cell phenotyping, subsequent spatial analysis, including metrics, signed distances and Voronoi diagrams, at single-cell resolution. It produces tables for fluorescence intensity quantification, binary positivity matrices, cell phenotype counts, and distance maps characterizing tumor heterogeneity, immune infiltration, and cell–cell proximity. Second, a U-Net model enhanced with Squeeze-and-Excitation attention blocks is implemented to detect invasion fronts in H&E-stained sections. Trained on over 2 000 semi-automatically annotated patches, it achieves a global F1 score of 0.82 and an AUC > 0.90. Applied to head and neck carcinomas, where this tool was rigorously validated, it revealed that approximately one-third of cytokeratin-positive epithelial cells express NGFR, with that subniche showing a 1.5× increase in proliferation (Ki-67⁺) and enrichment for PD-L1. Furthermore, CD8⁺ T cells infiltrate the stroma more readily than macrophages, while regulatory T cells preferentially localize to NGFR⁺ regions, suggesting a localized immunosuppressive microenvironment. Containerized and compliant with OME-TIFF and CLIA standards, this pipeline offers a scalable platform for discovering and validating simultaneously multiple biomarkers in translational research and diagnostics.