1Research Scholar, Sathyabama Institute of Science and Technology. Email: binnylatheesh@gmail.com
2Professor, Department of Computer Science & Engineering, School of Computing (Deemed to be University), Chennai-600 119. Email: psmaran@sathyabama.ac.in
Juvenile Rheumatoid Arthritis (JRA) represents a critical pediatric autoimmune condition requiring early intervention to prevent irreversible joint damage and systemic complications. Traditional diagnostic approaches suffer from delayed recognition due to atypical symptom presentations and reliance on subjective clinical assessments. This study introduces a novel hybrid deep learning-ensemble framework incorporating multi-modal feature fusion for automated JRA detection in adolescent populations aged 12-18 years. Our proposed methodology integrates clinical biomarkers, radiological imaging features, genetic predisposition indicators, and temporal symptom progression patterns through a sophisticated attention-based neural architecture combined with ensemble learning techniques.
The framework employs a three-stage pipeline: (1) multi-modal data preprocessing with advanced feature extraction using convolutional neural networks for imaging data and transformer architectures for sequential clinical measurements, (2) adaptive feature selection through genetic algorithm-optimized recursive feature elimination, and (3) hybrid classification using stacked ensemble methods combining XGBoost, LightGBM, and deep neural networks with uncertainty quantification. Experimental validation on a comprehensive dataset of 2,847 adolescent patients demonstrates superior performance with 94.3% accuracy, 92.7% sensitivity, and 95.8% specificity, significantly outperforming traditional machine learning approaches and existing clinical diagnostic protocols.
The proposed framework introduces several novel contributions including temporal biomarker trend analysis, multi-scale radiological feature extraction, and explainable AI components for clinical decision support. Real-world deployment simulations indicate potential for 40% reduction in diagnostic delays and 60% improvement in early intervention outcomes. This research establishes a new paradigm for AI-assisted pediatric rheumatology diagnosis with direct implications for precision medicine and personalized treatment strategies.
Keywords: Juvenile Rheumatoid Arthritis, Deep Learning, Multi-Modal Fusion, Ensemble Methods, Predictive Analytics, Biomarker Analysis, Explainable AI, Precision Medicine
How to cite this article: Binny S, Maran PS. Hybrid Deep Learning-Ensemble Framework with Multi-Modal Feature Fusion for Early Detection of Juvenile Rheumatoid Arthritis: A Novel Predictive Analytics Approach. Int J Drug Deliv Technol. 2026;16(11s): 51-58; DOI: 10.25258/ijddt.16.11s.6
Source of support: Nil.
Conflict of interest: None