Spatial Profiling Initiative

Advances in digital pathology have enabled unprecedented insights into tissue organization, tumor microenvironments, and disease progression through multiplex imaging and computational methods. However, challenges remain in scaling analyses to gigapixel whole-slide images, handling complex datasets, and ensuring reproducibility. We present an integrated framework combining three innovative approaches from our research lab: HTCNet, ImPartial, and SPT, to address these challenges. 

HTCNet introduces a hierarchical tissue-cell semantic segmentation model that efficiently infers dense tissue and cell annotations from H&E images using only partial labels. By leveraging a novel consistency loss function, HTCNet achieves state-of-the-art performance on benchmark datasets with reduced computational requirements and minimal deterioration under lower resolution or label coverage. 

ImPartial complements this by enabling interactive instance segmentation for whole-cell/nuclei/vessel using partial annotations on multiplex immunofluorescence images, incorporating human iterative feedback to refine segmentation boundaries and address variability in multiplex imaging datasets. Using active learning and self-supervised denoising, ImPartial achieves optimal segmentation performance with reduced annotation costs and training time. 

SPT extends these capabilities by providing a no-code spatial analytics platform for reproducible digital pathology. It enables real-time analysis of curated multiplexed pathology datasets, supported by versioned pipelines for data retrieval, preprocessing for use by models like ImPartial and HTCNet, and archiving. Findings are shareable via URLs and contribute to an open-contribution model, lowering barriers for scientific collaboration and dissemination. 

Together, these tools form an integrated approach to digital pathology, emphasizing reproducibility, scalability, and accessibility. Applications include spatial biomarker derivation, immunotherapy response prediction, and disease progression analysis, showcasing the transformative potential of computational pathology in therapeutic and diagnostic practice.