Biomarker profiling and integrating heterogeneous models for enhanced multi-grade breast cancer prognostication

Area of Science:

• Oncology
• Artificial Intelligence
• Biomarker Research

Background:

• Breast cancer is a leading cause of mortality in women globally.
• Limited awareness and screening resources hinder early diagnosis and effective treatment.
• Accurate, early detection is critical for improving survival rates.

Purpose of the Study:

• To develop an AI-based model for predicting breast cancer and its histopathological grades.
• To integrate multiple biomarkers and subject age for enhanced diagnostic accuracy and prognostication.
• To advance breast cancer screening and personalized treatment strategies.

Main Methods:

• An ensemble machine learning framework integrating beta-human chorionic gonadotropin (β-hCG), Programmed Cell Death Ligand 1 (PD-L1), and alpha-fetoprotein (AFP) with subject age.
• Particle Swarm Optimization (PSO) for hyperparameter tuning and minority oversampling to prevent overfitting.
• Model performance validated using five-fold cross-validation.

Main Results:

• The AI model achieved 97.93% accuracy and 98.06% F1-score on test data.
• Demonstrated superior performance compared to state-of-the-art methods across diverse age groups.
• Highlighted robustness and generalizability of the proposed framework.

Conclusions:

• The study presents a significant advancement in breast cancer screening by integrating multiple biomarkers for accurate tumor grading.
• The AI framework offers potential to reduce mortality rates, especially in resource-limited settings.
• Enables improved early intervention and personalized treatment strategies for breast cancer patients.