Qiqi Kong, Yunqing Ban
Tumor heterogeneity is a core challenge in gynecologic oncology, driving therapeutic resistance and limiting the efficacy of single-point biopsies. Artificial intelligence (AI) and radiomics are emerging as a "digital biopsy" to non-invasively decode tumor biology from medical radiological modalities images(including MRI, CT, and PET). This review synthesizes the state of AI in predicting key molecular features across gynecologic cancers, including homologous recombination deficiency (HRD) in ovarian cancer, microsatellite instability (MSI) and PI3K activation in endometrial cancer, and, as an illustrative case, HPV integration and DNA methylation in cervical cancer. We further explore how advanced architectures like Vision Transformers (ViTs) and Graph Neural Networks (GNNs) can delineate the tumor microenvironment and predict therapeutic response. Finally, we discuss critical hurdles to clinical translation-such as model generalizability, the need for causal AI, and the data bottleneck-while examining future paradigms like foundation models and patient-specific "digital twins." This review highlights AI's revolutionary potential to link imaging phenotype with molecular genotype, advancing a new era of precision medicine in gynecologic oncology.