Unlocking Precision Medicine: A Fully Automated Multi-Modal CTCA Based System For Risk Prediction

Introduction
Understanding major adverse cardiac events (MACE) risk is fundamental to improving cardiovascular health. We explored enhancement in risk prediction by integrating a fully automated Coronary Artery Disease Reporting and Data System (CAD-RADS) with patient demographics and detailed anatomical data from Coronary Computed Tomography Angiography (CTCA) scans using a multi-modal deep learning system.
Methods
We employed convolutional neural networks for automated CAD-RADS generation and a gradient-boosting decision tree model to evaluate CAD-RADS' effectiveness in predicting MACE. We then built a multi-modal deep learning system that combined automated CAD-RADS with patient demographics and CTCA-derived segmentation of the left ventricle, aorta, and heart. We evaluated the performance of different models (i.e., expert-generated and fully automated CAD-RADS, and the multimodal model system), using the area under the curve (AUCROC).
Results
Among 995 patients, 639 with both imaging and clinical data (mean age 69.9±8.7 years, 361 males) were studied. Within 30 days, 45 patients experienced MACE. Automated CAD-RADS (AUCROC=0.69) demonstrated comparable performance to expert human assessments (AUCROC =0.67, p-value=0.77), while the multi-modal DL system (AUCROC = 0.821) outperformed CAD-RADS in predicting MACE (p-value=0.02), achieving better sensitivity (0.78) and specificity (0.79).
Conclusions
The novel multi-modal system built using fully automated CAD-RADS and CTCA-derived segmentations, along with patient demographics; outperforms both the human expert-generated and fully automated CAD-RADS for MACE prediction. This approach has the potential to enhance patient outcomes by leveraging the synergy between automated imaging assessments and comprehensive patient data.