Resolving phenotyping discordance with SPACEMAP, an integrated machine learning framework

Abstract

Multiplex imaging technologies have revolutionized our ability to study cellular behavior within the tissue microenvironment. Translating this complex data into meaningful biological insights requires a unified analytical framework. To address this, we developed SPACEMAP (Spatial Phenotyping And Classification with Enhanced Multiplex Analysis Pipeline), a comprehensive Python and Qupath-based platform for multiplex imaging analysis. SPACEMAP integrates image registration, segmentation, artifact removal, tissue and zone classification, spatial feature extraction, and a consolidated phenotyping approach into a single system. A core feature of SPACEMAP is its high-fidelity phenotyping. To evaluate classification performance, we benchmarked our method RESOLVE, against three established approaches, Leiden clustering, Self-Organizing Maps, and SCIMAP revealing substantial disagreement among them. SPACEMAP overcomes this through two complementary workflows: a machine learning model trained on expert-labeled cells, and a consensus classifier that integrates high-confidence cells across methods. Here, we validated SPACEMAP on in-house colorectal cancer samples and a public dataset, demonstrating its robustness.