Exploring the Distribution of Cell Subpopulations in Pancreatic Ductal Adenocarcinoma Slides by Joint Spatial Transcriptomics and Pathology Data

Current spatial transcriptomics (ST) technology can integrate stained pathological slides with RNA sequencing, providing precise information on gene expression and cell types. However, the unique handling of pathological slides required by ST technology can lead to image quality issues, impacting model performance. To address this challenge, we applied Structure Preserving Color Normalization (SPCN), based on the Beer-Lambert law. Here, we use a pancreatic ductal adenocarcinoma (PDAC) dataset with class-level and gene-level annotations, and introduced Gene Feature Align (GFA) Loss to align classification features with gene expression, resulting in clearer decision boundaries. Our approach significantly enhances model performance on this dataset, offering novel solutions for spatial transcriptomics and paired pathological images, while contributing to limited research on PDAC slide cell distribution prediction through deep neural networks.