Artificial intelligence in endometrial cancer: multimodal deep learning and future perspectives in precision gynecologic oncology - a structured narrative review.
Authors
Affiliations (7)
Affiliations (7)
- Department of Gynecology and Obstetrics, Faculty Hospital Nitra, Nitra, Slovakia.
- Faculty of Social Sciences and Health Care, Constantine the Philosopher University in Nitra, Nitra, Slovakia.
- Department of Health Studies, The College of Polytechnics Jihlava, Jihlava, Czechia.
- Department of Oncology, Faculty Hospital Nitra, Nitra, Slovakia.
- Department of Oncology, First Faculty of Medicine of Charles University and General University Hospital in Prague, Prague, Czechia.
- Department of Laboratory Medicine, Faculty Hospital Nitra, Nitra, Slovakia.
- 1st Department of Gynecology and Obstetrics, Faculty of Medicine, Comenius University Bratislava, Bratislava, Slovakia.
Abstract
Endometrial cancer (EC) is the most common gynecologic malignancy in developed countries, with a rising incidence driven by obesity, metabolic syndrome, diabetes mellitus, and aging populations. Artificial intelligence (AI) has rapidly advanced in gynecologic oncology, particularly in diagnostic imaging, digital pathology, radiomics, and multimodal prognostic modeling. Recent breakthroughs in machine learning, deep learning, and whole-slide imaging (WSI), including multimodal prognostic architectures such as HECTOR and explainable AI (XAI), have enabled the development of sophisticated diagnostic and prognostic systems. These systems can differentiate benign from malignant lesions, predict myometrial invasion and lymph node metastases, stratify recurrence risk, and support individualized therapeutic decision-making. Notably, recent multimodal deep-learning models, particularly the HECTOR framework, have demonstrated improved prognostic performance compared with conventional clinicopathologic risk-stratification systems for predicting distant recurrence in externally validated cohorts, although further prospective validation remains necessary. Despite these promising advances, current evidence remains limited by predominantly retrospective study designs, relatively small and frequently imbalanced datasets, potential overfitting of AI models, heterogeneous imaging protocols, limited external validation, and insufficient prospective clinical evidence. These limitations restrict model generalizability across different populations and healthcare settings. Furthermore, ethical, regulatory, and data-governance challenges remain important barriers to widespread clinical implementation. This structured narrative review summarizes AI applications in endometrial cancer, focusing on ultrasound imaging, radiomics, digital pathology, multimodal deep-learning, explainable AI, prognostic modeling, clinical translation, and future perspectives in personalized care.