Spatially-Delineated Domain-Adapted AI Classification: An Application for Oncology Data
Majid Farhadloo, Arun Sharma, Alexey Leontovich, Svetomir N. Markovic, Shashi Shekhar
TL;DR
The paper tackles unsupervised domain adaptation for classifying target place-types when spatial variability across multi-type point maps complicates transfer learning. It introduces a spatially-delineated AI classifier built as a multi-task framework that couples supervised learning on a labeled source with spatially-aware self-supervised tasks—spatial mix-up masking (SMUM) and spatial masking—and a spatial contrastive predictive coding (SCPC) objective to capture spatial arrangements shared across place-types. Empirical results on real MxIF oncology data show the approach outperforming adversarial and geometry-aware baselines, with architecture choice (DGCNN vs SAMCNet) and the SSL tasks significantly affecting gains. A case study demonstrates clinically meaningful spatial interactions in the tumor microenvironment, supporting the method’s potential to assist pathologists and inform immunotherapy design. Overall, the work advances spatially informed domain adaptation for complex biomedical point-map data and outlines directions for integrating domain knowledge and privacy-preserving data generation.
Abstract
Given multi-type point maps from different place-types (e.g., tumor regions), our objective is to develop a classifier trained on the source place-type to accurately distinguish between two classes of the target place-type based on their point arrangements. This problem is societally important for many applications, such as generating clinical hypotheses for designing new immunotherapies for cancer treatment. The challenge lies in the spatial variability, the inherent heterogeneity and variation observed in spatial properties or arrangements across different locations (i.e., place-types). Previous techniques focus on self-supervised tasks to learn domain-invariant features and mitigate domain differences; however, they often neglect the underlying spatial arrangements among data points, leading to significant discrepancies across different place-types. We explore a novel multi-task self-learning framework that targets spatial arrangements, such as spatial mix-up masking and spatial contrastive predictive coding, for spatially-delineated domain-adapted AI classification. Experimental results on real-world datasets (e.g., oncology data) show that the proposed framework provides higher prediction accuracy than baseline methods.