Stacking Machine Learning Algorithms for Biomarker-Based Preoperative Diagnosis of a Pelvic Mass.

Area of Science:

• Gynecologic Oncology
• Biomarker Discovery
• Machine Learning in Medicine

Background:

• Pelvic masses present a diagnostic challenge, requiring accurate preoperative differentiation between benign and malignant conditions.
• Early and precise diagnosis of ovarian malignancy is crucial for effective treatment planning and improved patient outcomes.

Purpose of the Study:

• To identify the most predictive parameters for ovarian malignancy.
• To develop a machine learning (ML) based algorithm for preoperative distinction between benign and malignant pelvic masses.

Main Methods:

• Retrospective analysis of 70 parameters from 140 women with pelvic masses.
• Feature selection using random forest and principal component analysis; evaluation of nine ML classifiers.
• Ensemble stacking of ML models via LASSO regression for enhanced predictive performance.

Main Results:

• HE4, CA125, and transferrin were identified as the top three predictive parameters for malignancy.
• The ensemble ML model achieved 97.1% accuracy, 0.951 AUC, 93.3% sensitivity, and 100% specificity on the testing dataset.
• The ensemble stack significantly outperformed individual ML classifiers in predicting malignancy.

Conclusions:

• Combining HE4, CA125, and transferrin measurements with an ensemble ML approach offers robust preoperative diagnostic capabilities for pelvic masses.
• This integrated biomarker and ML strategy can significantly aid in the preoperative assessment of ovarian malignancy.
• The developed ensemble model shows high potential for clinical application in improving the diagnostic accuracy of pelvic masses.