Integrating artificial intelligence and multi-omics data for precision oncology in endometrial cancer: a narrative...

Endometrial cancer (EC) is the most common gynaecological malignancy worldwide, yet the prognosis for advanced and recurrent disease remains poor, highlighting the need for improved diagnostic, prognostic, and therapeutic decision-making frameworks. Conventional approaches, including histopathology, imaging, and single-layer molecular profiling, provide essential clinical information but may not fully capture EC's biological heterogeneity, especially within clinically challenging No Specific Molecular Profile (NSMP) and mismatch repair-deficient (MMRd) subgroups. Artificial intelligence (AI) and machine learning (ML) provide powerful approaches to analyse complex, high-dimensional datasets generated by multi-omics profiling, histopathology, imaging, and clinical records.In this review, we synthesize the latest evidence on AI-driven multi-omics research in EC, encompassing genomics, transcriptomics, proteomics, metabolomics, epigenomics, single-cell profiling, and spatial transcriptomics. Unlike other reviews that focus solely on AI, omics, or imaging, we integrate molecular, imaging, histopathological, and computational perspectives to underscore their collective impact on precision oncology in EC. We subsequently explore applications in molecular subtyping, predicting survival and recurrence, modelling treatment responses, discovering immunotherapy biomarkers, and identifying drug targets. Public resources such as The Cancer Genome Atlas (TCGA), Clinical Proteomic Tumour Analysis Consortium (CPTAC), Gene Expression Omnibus (GEO), cBioPortal, Human Protein Atlas, GTEx, and UCSC Xena have enabled large-scale reproducible analyses. However, challenges such as cohort heterogeneity, batch effects, ethnic underrepresentation, missing annotations, and the need for external validation remain significant hurdles.We then discuss the progression from conventional ML methods to deep learning architectures, including convolutional neural networks, transformers, graph neural networks, and multimodal fusion models applied to histopathological, radiological, and multi-omics data. Landmark models such as EndoNet, EndoRisk, and HECTOR illustrate the potential of AI-enabled approaches to support EC grading, molecular inference, lymph node metastasis prediction, and recurrence-risk stratification. Finally, we examine key translational barriers, including class imbalance, interpretability, data harmonization, regulatory requirements, and the implementation gap between high-performing retrospective models and routine clinical deployment. Ultimately, this review underscores how bridging these multi-modal computational approaches paves the way for precision oncology in EC.