BIDCell is a self-supervised deep learning framework designed to segment subcellular spatial transcriptomics (SST) data. It incorporates biologically informed loss functions and cell-type data from single-cell transcriptomics datasets to learn relationships between spatial gene expression and cell morphology. The framework addresses the challenges of cell segmentation in SST images by leveraging high-dimensional expression information and prior knowledge from single-cell sequencing data. BIDCell outperforms existing methods in terms of cell segmentation performance, as demonstrated through a comprehensive evaluation framework called CellSPA, which assesses five complementary categories of metrics. BIDCell's ability to capture diverse cell shapes, improve expression purity, and reduce contamination makes it a promising tool for enhancing single-cell spatial expression analyses and biological discovery. The method's flexibility and generalizability across different SST platforms and tissue types further highlight its potential impact on the field of subcellular spatial transcriptomics.BIDCell is a self-supervised deep learning framework designed to segment subcellular spatial transcriptomics (SST) data. It incorporates biologically informed loss functions and cell-type data from single-cell transcriptomics datasets to learn relationships between spatial gene expression and cell morphology. The framework addresses the challenges of cell segmentation in SST images by leveraging high-dimensional expression information and prior knowledge from single-cell sequencing data. BIDCell outperforms existing methods in terms of cell segmentation performance, as demonstrated through a comprehensive evaluation framework called CellSPA, which assesses five complementary categories of metrics. BIDCell's ability to capture diverse cell shapes, improve expression purity, and reduce contamination makes it a promising tool for enhancing single-cell spatial expression analyses and biological discovery. The method's flexibility and generalizability across different SST platforms and tissue types further highlight its potential impact on the field of subcellular spatial transcriptomics.