Endometrial cancer in women with abnormal uterine bleeding: Data mining classification methods

Area of Science:

• Medical Artificial Intelligence
• Oncology
• Machine Learning in Healthcare

Background:

• Growing application of artificial intelligence (AI) in medicine.
• Early diagnosis of endometrial cancer is crucial for effective treatment.
• Need for non-invasive screening methods, particularly for high-risk individuals.

Purpose of the Study:

• Investigate risk factors associated with endometrial cancer.
• Develop a machine learning tool for endometrial cancer prediction.

Main Methods:

• Cross-sectional study of 972 patients with abnormal uterine bleeding (January 2016 - January 2021).
• Analysis of patient characteristics and curettage pathology findings.
• Application of machine learning algorithms: artificial neural networks, classification and regression trees, support vector machine, and logistic regression.

Main Results:

• Out of 972 patients, 52 were diagnosed with endometrial cancer.
• Logistic regression model achieved the highest prediction performance (100% sensitivity, 98.83% specificity on trained data; 98% sensitivity, 98.7% specificity on test data).
• Classification and regression trees showed the next best performance.

Conclusions:

• AI-based algorithms show promise as a non-invasive screening method for endometrial cancer.
• Machine learning models, particularly logistic regression, can effectively predict endometrial cancer risk.