Spatial transcriptomics has emerged in recent years as an advanced technique integrating modern microscopy and single-cell RNA sequencing (RNA-seq). The computational aspect of spatial transcriptomics analysis is actively evolving, with software packages like Squidpy at the forefront. A critical challenge in this field is achieving reliable image segmentation based on RNA expression levels within cells, necessitating robust sets of genomic signatures. In this paper, we present our approach to addressing this challenge through data integration and by leveraging proteome signatures from The Human Protein Atlas as a reference standard, applicable to transcriptomic profiles. Our heuristic-based segmentation method provides biological validation for unsupervised techniques and serves as a preliminary proof of concept. Additionally, it establishes a benchmark for the future application of artificial intelligence with the use of transcriptome foundation models.