PatternExtract: A Facile, Scalable Pipeline for Point Pattern Generation from Spatial Imaging Data.

Abstract

Spatial point pattern analysis offers a mathematically rigorous framework for characterizing cell distribution in cancer tissue, yet existing pipelines lack robust methods for defining accurate spatial windows that exclude noncellular artifacts. We present PatternExtract, an open-source, multi-platform pipeline for generating biologically accurate spatial point patterns from multiplexed imaging data. The pipeline introduces a 2-kernel tissue segmentation approach that overlays concentric circular kernels on cellular coordinates to construct tissue masks without dependence on proprietary software or fluorescence composite channels, making it compatible with RGB images and coordinate outputs from any cell segmentation tool. Benchmarked on 568 images from 274 diffuse large B cell lymphoma (DLBCL) patients, PatternExtract identified and correctly excluded tissue artifacts in 27% of images, reducing window area by a mean of 9.8 ± 8.2% relative to convex hull methods. Convex hull window misspecification inflated K function AUC by up to 649.6% and produced 100% false rejection of complete spatial randomness under simulation, errors fully corrected by PatternExtract. Cross-cohort validation on an independent 20-patient cohort confirmed pipeline consistency across imaging platforms and staining protocols. Biological application to Ki67⁺ cell distributions in DLBCL demonstrated the pipeline's utility in detecting genuine short-range clustering beyond spatial inhomogeneity. PatternExtract is available at https://github.com/shrutisridhar99/PatternExtract with a Streamlit-based graphical user interface (GUI) for accessibility across programming backgrounds.