Tyramide-Amplified Protein Imaging Enables Boundary-Resolved Cell Segmentation for Improved Spatial Transcriptomic Analysis

Many spatial biology workflows rely on accurate delineation of full-cell boundaries, yet commonly used segmentation strategies estimate cellular extent indirectly from nuclear morphology or spatial transcript distributions. These approaches can perform poorly in densely packed tissues or regions containing cells with irregular morphologies, leading to inaccurate transcript assignment and errors in downstream spatial transcriptomic analysis. Here, we developed TSA-Seg, a cell segmentation framework fine-tuned on tyramide signal amplification-labeled protein images that provide continuous morphological information across whole cells. Unlike DAPI-Seg, which relies on fixed-radius nuclear expansion, TSA-Seg generates full-cell masks directly from the TSA-amplified protein morphology. Across cultured cells and formalin-fixed paraffin-embedded human tissue sections, TSA-Seg improved segmentation accuracy, particularly in crowded and morphologically complex regions. TSA-Seg was also generalized to multiple tissue types and protein markers not used during model fine-tuning. In targeted in situ sequencing datasets from non-small-cell lung cancer tissues, TSA-Seg produced more morphology-consistent boundaries than RNA-based segmentation. Across 46 analyzed tissue regions, TSA-Seg reduced the mean transcript assignment error rate from 47% with DAPI-Seg to 16% and yielded more spatially coherent localization of tumor- and immune-associated transcripts relative to the predicted cell boundaries. These findings establish TSA-amplified protein morphology as an experimentally accessible boundary readout for improving cell-level spatial transcriptomic analysis in complex tissues.