From image biomarkers to treatment decisions: multimodal AI and clinical translation in gynecologic oncology

Artificial intelligence (AI) is rapidly expanding across gynecologic oncology, but most applications remain focused on image interpretation and retrospective prediction rather than direct support of treatment decisions. This review examines how AI is progressing from imaging biomarkers toward multimodal decision-support systems that integrate radiologic, pathologic, molecular, and clinical information. We summarize current applications in ovarian, endometrial, and cervical cancers, with particular attention to preoperative risk stratification, surgical planning, radiotherapy, systemic treatment selection, treatment-response prediction, and multidisciplinary decision-making. Multimodal approaches are especially promising because they reflect the way clinical decisions are made and may provide greater value than models based on a single data source. However, the current evidence remains dominated by retrospective and often single-center studies, with limited external and prospective validation. Data heterogeneity, domain shift, incomplete multimodal datasets, bias, workflow integration, regulatory requirements, and model monitoring remain important barriers to routine implementation. We therefore outline a stepwise pathway for clinical translation, from robust multicenter validation and prospective human–AI evaluation to clinical-impact studies and post-deployment monitoring. The next phase of AI in gynecologic oncology should prioritize clinically meaningful decisions and measurable patient or workflow benefit rather than incremental gains in predictive performance alone.