CERV-Score: A Hybrid Machine Learning Framework for Cervical Cancer Risk Prediction Using Integrated Clinical and...

Area of Science:

• Oncology
• Genomics
• Machine Learning

Background:

• Cervical cancer poses a significant global health challenge, with late diagnoses leading to high mortality, especially in underserved areas.
• Accurate early risk prediction is crucial for improving patient outcomes and guiding preventive strategies.

Purpose of the Study:

• To introduce CERV-Score, a novel hybrid machine learning framework for enhanced cervical cancer risk prediction.
• To develop a clinical-genomic decision support tool for real-time, actionable risk assessment.

Main Methods:

• Developed a hybrid machine learning model (CERV-Score) integrating clinical risk factors and recurrence-based genomic markers.
• Utilized recurrence analysis of RNA-seq data to identify robust genomic markers.
• Created an interactive decision support tool with probabilistic risk scoring and gene lookup functionality.

Main Results:

• The CERV-Score model demonstrated high predictive performance with 94.1% accuracy, an F1-score of 0.91, and an AUC of 0.94.
• Bootstrap resampling confirmed the model's robust performance with a 95% confidence interval for accuracy (92.8%-95.4%).
• The model enables nuanced patient stratification into low, moderate, and high-risk groups.

Conclusions:

• CERV-Score offers a significant advancement in cervical cancer risk prediction through probabilistic scoring and genomic integration.
• The hybrid approach enhances accuracy, interpretability, and clinical utility for personalized patient management.
• This framework lays the foundation for more deployable and interpretable cervical cancer risk prediction systems.