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

• Published in: Computer methods and programs in biomedicine Vol. 255; p. 108349
• Main Authors: Joshi, Rakesh Chandra, Srivastava, Pallavi, Mishra, Rashmi, Burget, Radim, Dutta, Malay Kishore
• Format: Journal Article
• Language: English
• Published: Ireland Elsevier B.V 01.10.2024
• Subjects: AI-based screening; Biomarker-driven classification; Breast cancer diagnosis; Ensemble learning; Machine learning in healthcare; Multi-grade prediction; Multi-grade prediction; Ensemble learning; AI-based screening; Machine learning in healthcare; Biomarker-driven classification; Breast cancer diagnosis
• ISSN: 0169-2607; 1872-7565; 1872-7565
• DOI: 10.1016/j.cmpb.2024.108349

•AI breast cancer prediction integrating β-hCG, PD-L1 and AFP biomarkers with age.•Heterogeneous models integration for enhanced Multi-Grade Breast Cancer prediction.•Hyperparameter fine-tuning via particle swarm optimization ensures model efficacy.•Cross-validation is done to analyse the outcomes obtained from trained models.•Potential to reduce breast cancer mortality rates with AI-enhanced screening. Breast cancer remains a leading cause of female mortality worldwide, exacerbated by limited awareness, inadequate screening resources, and treatment options. Accurate and early diagnosis is crucial for improving survival rates and effective treatment. This study aims to develop an innovative artificial intelligence (AI) based model for predicting breast cancer and its various histopathological grades by integrating multiple biomarkers and subject age, thereby enhancing diagnostic accuracy and prognostication. A novel ensemble-based machine learning (ML) framework has been introduced that integrates three distinct biomarkers—beta-human chorionic gonadotropin (β-hCG), Programmed Cell Death Ligand 1 (PD-L1), and alpha-fetoprotein (AFP)—alongside subject age. Hyperparameter optimization was performed using the Particle Swarm Optimization (PSO) algorithm, and minority oversampling techniques were employed to mitigate overfitting. The model's performance was validated through rigorous five-fold cross-validation. The proposed model demonstrated superior performance, achieving a 97.93% accuracy and a 98.06% F1-score on meticulously labeled test data across diverse age groups. Comparative analysis showed that the model outperforms state-of-the-art approaches, highlighting its robustness and generalizability. By providing a comprehensive analysis of multiple biomarkers and effectively predicting tumor grades, this study offers a significant advancement in breast cancer screening, particularly in regions with limited medical resources. The proposed framework has the potential to reduce breast cancer mortality rates and improve early intervention and personalized treatment strategies.