INTEGRATION OF HISTOPATHOLOGY AND SPATIAL TRANSCRIPTOMICS FOR CANCER CLASSIFICATION: A SYSTEMATIC REVIEW

Abstract

Background: Conventional histopathology remains fundamental to cancer diagnosis, but morphology alone cannot fully resolve intratumoral molecular heterogeneity. Spatial transcriptomics (ST) preserves gene expression information within tissue architecture and provides a route to connect microscopic phenotype with molecular state. Recent artificial-intelligence methods combine hematoxylin-and-eosin (H&E) images with spatial transcriptomic profiles to predict gene expression, classify tissue domains, delineate molecular subtypes, and characterize tumor–immune and clonal heterogeneity. Methods: MEDLINE/PubMed, Embase, Scopus, Web of Science, and IEEE Xplore were searched from January 2018 through 30 June 2026. Citation and reference-list searching supplemented database retrieval. Eligible studies analyzed human cancer tissue and jointly incorporated histopathology with spatially resolved transcriptomic information for cancer classification, tissue-domain identification, molecular subtyping, cell-state assignment, clonal mapping, or spatial gene-expression prediction relevant to cancer classification. Because of heterogeneity in cancer types, ST platforms, model architectures, target genes, spatial resolution, and performance metrics, a narrative synthesis was performed. Results: A total of 2,756 records were identified: 2,682 from databases and 74 through citation/reference searching. After removal of 689 duplicates, 2,067 records were screened and 1,781 excluded. Of 286 reports sought for retrieval, 17 were unavailable. Two hundred sixty-nine full texts were assessed and 243 excluded, leaving 26 studies for qualitative synthesis. Breast cancer was the most frequently represented malignancy. Across the evidence base, convolutional, transformer, graph-neural-network, contrastive-learning, and foundation-model approaches increasingly linked morphology with spatial gene expression and biologically meaningful tumor domains. Independent benchmarking showed that no single method dominated accuracy, generalizability, computational efficiency, and translational performance. Direct clinically oriented cancer classification endpoints remained less common than gene-expression prediction. Conclusions: Integration of histopathology and spatial transcriptomics can enrich cancer classification by linking morphology with molecular and spatial phenotypes. Evidence is strongest for spatial gene-expression inference and tissue-domain characterization, while direct multicenter validation of molecular subtype, prognosis, and treatment-response classification remains limited. Standardized benchmarking, patient-level external validation, and prospective clinical evaluation are required before routine diagnostic adoption.