Endometrial cancer (EC) is one of the most common cancers affecting women, with an 83% five-year survival rate when diagnosed. It is particularly prevalent in developed countries like Europe and North America, partly due to sedentary lifestyles and genetic factors.
EC is traditionally divided into two types based on hormonal dependence. Type 1 is estrogen-dependent and tends to grow in response to elevated estrogen levels. Type 2 is estrogen-independent and often carries a worse prognosis regardless of hormone status.
A newer approach classifies EC based on genomic and molecular alterations alongside biological markers. This molecular classification has shown greater accuracy in predicting how individual patients will respond to treatment, enabling more tailored care.
Treatment options include surgery, chemotherapy, and radiation therapy. The standard surgical approach is a total hysterectomy with bilateral salpingo-oophorectomy (removal of the uterus, ovaries, and fallopian tubes), sometimes combined with lymph node removal depending on disease stage.
Accurate diagnosis of endometrial cancer is challenging because histological (tissue-based) criteria vary and pathologists can disagree on interpretation. This variability can lead to either undertreatment (missing a cancer) or overtreatment (unnecessary intervention), both of which have serious consequences for patients.
The standard diagnostic tools include transvaginal ultrasound (TVUS) and MRI. TVUS is operator-dependent and requires skilled practitioners, while MRI is more specific but only available in tertiary medical centers. Both provide limited qualitative information about tumor characteristics.
Endometrial biopsy, guided by hysteroscopy, is considered the gold standard for tissue diagnosis. The Pipelle device - a simple aspiration tool invented in 1984 - is commonly used as a less invasive alternative, though it has a meaningful failure rate in obtaining sufficient tissue samples.
A critical diagnostic challenge is distinguishing atypical endometrial hyperplasia (AEH) - an abnormal but pre-cancerous thickening of the uterine lining - from true malignancy. Up to 25% of women with AEH may already have a concurrent well-differentiated cancer, making accurate differentiation essential for treatment planning.
Artificial intelligence (AI) in medicine broadly refers to computer systems that can analyze medical data and assist in clinical decisions. It encompasses two main areas: a virtual component (machine learning and deep learning) and a physical component (robotic systems and care bots).
Deep learning is a subset of AI that uses artificial neural networks - computational systems modeled loosely on the human brain. These networks process massive datasets, identify patterns, and improve their accuracy with each iteration, without being explicitly programmed for every scenario.
Computer vision is a key application of AI in medicine, allowing computers to analyze medical images - such as MRI scans, CT scans, pathology slides, and ultrasounds - in a manner comparable to trained human radiologists and pathologists.
In radiology, AI has shown the most success to date, helping reduce workload, minimize errors from inexperienced practitioners, and speed up diagnosis in emergency settings. AI systems have been validated for detecting pathologies in the breast, lungs, liver, and bones, though human oversight remains essential.
Zhang and colleagues applied deep learning to hysteroscopic images (images captured inside the uterus) to classify endometrial lesions as cancerous or benign. Training on over 6,400 images from 454 patients, their model matched or exceeded the diagnostic accuracy of gynecologists, demonstrating that AI can meaningfully assist in real-time tissue evaluation.
Zhao et al. identified 14 key genes involved in EC and built an artificial neural network (ANN) diagnostic model using public genomic databases (GEO and TCGA). This model showed high sensitivity and specificity for detecting early-stage EC, pointing to the value of genetic data as an input for AI-driven diagnosis.
Research using AI-enhanced MRI with convolutional neural networks showed high diagnostic performance in identifying EC preoperatively. Since MRI typically only produces qualitative descriptions, AI's ability to extract quantitative image features dramatically improves diagnostic precision.
Hart and colleagues tested seven different AI algorithms for population-level EC screening and found that a random forest algorithm performed best. This approach was both non-invasive and cost-effective compared to traditional screening methods, suggesting AI could expand screening access to underserved populations.
FIGO staging - developed by the International Federation of Gynaecologists and Obstetricians - classifies EC from Stage I (confined to the uterus) through Stage IV (distant metastasis). AI models have been developed to assist in automating and improving the accuracy of this staging process from imaging and pathology data.
Feng et al. developed a deep learning model trained on hematoxylin and eosin (H&E) stained histopathological slides that could predict lymph node metastasis - a critical factor in staging and treatment decisions. The model showed enhanced accuracy, especially in early-stage patients where metastasis prediction is most impactful.
Robotic-assisted surgery, a physical form of AI-assisted intervention, has shown significant advantages: less blood loss, shorter hospital stays, fewer lymph nodes removed, and comparable survival rates to traditional laparoscopic surgery. These benefits make it an appealing option for EC staging and treatment.
A study by Kim et al. used CT-based AI to measure waist skeletal muscle volume, identifying it as a promising biomarker for predicting EC prognosis. Lower muscle mass (sarcopenia) was associated with worse outcomes, highlighting how AI can extract prognostic information from routine imaging.
Radiomics - the extraction of large amounts of quantitative features from medical images - combined with AI is an emerging approach. Preliminary data suggests radiomics can characterize tumors in ways invisible to the naked eye, potentially revolutionizing preoperative assessment.
A 2021 systematic review by Akazawa et al. analyzed 71 AI studies in gynecologic cancers, including 13 focused on EC. Studies used both imaging data (MRI, CT, ultrasound, hysteroscopy) and value-based data (blood tests, tumor markers, patient backgrounds). The primary targets were definitive diagnosis and prognostic outcomes like overall survival and lymph node metastasis.
A major finding was that 90% of studies included fewer than 1,000 cases, with a median dataset size of just 214 cases. Small datasets limit the generalizability of AI models and make it difficult to confirm their real-world reliability. Lack of external validation datasets further limits confidence in the findings.
A 2023 study by Erdemoglu et al. used multiple AI algorithms - including random forest, logistic regression, and neural networks - to predict EC risk in both pre- and post-menopausal women. Using age, BMI, endometrial thickness, and other clinical features, the best models achieved 94% accuracy with an AUC of 0.938, identifying high-risk women effectively.
Despite progress, cervical cancer has received far more AI research attention than endometrial cancer. This gap reflects a need for larger, multi-institutional studies specifically targeting EC to move the field forward.
The application of AI in healthcare raises significant ethical concerns around data privacy, algorithmic bias, transparency, and accountability. Machine learning systems trained on non-representative datasets can inadvertently encode and amplify existing health disparities, disadvantaging certain patient populations.
A central ethical challenge is the "black box" problem: many AI algorithms make predictions whose internal reasoning is opaque - not fully understandable even to their developers. This lack of interpretability makes it difficult for physicians and patients to assess how a decision was reached, raising concerns about informed consent and clinical responsibility.
The question of who bears legal responsibility when an AI system causes harm remains unresolved. Current frameworks suggest humans (clinicians and developers) retain accountability, but the complexity of AI decision-making challenges traditional models of medical liability.
Frameworks for responsible AI in healthcare must address four pillars: informed consent for data use, safety and transparency, algorithmic fairness, and data privacy. A multidisciplinary approach involving policymakers, developers, clinicians, and patients is needed to implement AI responsibly.
AI holds substantial promise for transforming endometrial cancer care across the entire clinical pathway - from screening and early diagnosis to treatment planning, surgical assistance, and ongoing monitoring of treatment response.
Machine learning algorithms can process patient records, imaging studies, and genomic data simultaneously to build personalized predictive models that identify high-risk individuals and suggest optimal treatment strategies based on disease subtype and patient characteristics.
For women with limited access to specialist healthcare, AI-powered screening tools could provide a cost-effective, non-invasive first line of risk assessment, potentially enabling earlier diagnosis in populations where the disease is disproportionately deadly due to late detection.
Fully realizing AI's potential will require larger, more diverse datasets, external validation studies, standardized reporting, and an ethical framework that keeps patient welfare at the center. The field is still in its infancy, and collaborative global research is essential to advance it responsibly.