AI-Driven Radiology: Multi-Modal Imaging Diagnosis Using Ensemble Models

Abstract: The integration of artificial intelligence (AI) in radiology has significantly improved diagnostic workflows, particularly with the advent of multi-modal imaging systems. This study proposes an ensemble deep learning framework for radiological diagnosis by combining complementary information from computed tomography (CT), magnetic resonance imaging (MRI), and positron emission tomography (PET). Modality-specific convolutional neural networks—ResNet50 for CT, DenseNet121 for MRI, and EfficientNet for PET—were independently trained and their outputs aggregated via a fusion layer for final classification. The models were evaluated using a stratified five-fold cross-validation approach on annotated multi-modal datasets, and performance was assessed using accuracy, precision, recall, F1-score, and AUROC.Results show that the ensemble model consistently outperformed individual modality models across all metrics, achieving an accuracy of 0.92 and an AUROC of 0.94. In comparison, single-modality models yielded AUROC scores ranging from 0.87 to 0.89. The ensemble approach demonstrated enhanced diagnostic performance and robustness, particularly in complex cases involving overlapping imaging features. These findings support the clinical viability of ensemble models for high-stakes diagnostic applications and highlight the benefits of integrative multi-modal learning in medical imaging. Keywords: AI in radiology, ensemble learning, multi-modal imaging, deep learning, CT, MRI, PET, diagnostic accuracy, convolutional neural networks, medical image analysis