UncertaintyFuseNet

UncertaintyFuseNet integrates chest computed tomography (CT) and X-ray imaging using a hierarchical feature-fusion deep learning model with Ensemble Monte Carlo Dropout to detect COVID-19 and quantify prediction uncertainty.

Key Features:

• Hierarchical Feature Fusion: Employs a hierarchical feature fusion approach to combine features extracted from chest computed tomography (CT) and X-ray images for joint classification.
• Uncertainty Quantification: Incorporates prediction uncertainty estimation using Ensemble Monte Carlo Dropout (EMCD) to provide confidence measures for outputs.
• Performance Metrics: Demonstrates 99.08% precision for CT scans and 96.35% accuracy for X-ray datasets, with evaluation using F-Measure and ROC curves.
• Robustness to Noise: Maintains performance on noisy and previously unseen data, indicating resilience to input variability.

Scientific Applications:

• COVID-19 Detection: Classifies COVID-19 cases from chest CT and X-ray images for automated diagnostic analysis.
• Computer-Aided Detection (CAD): Provides uncertainty-aware predictions to inform CAD systems applied to diverse and noisy imaging datasets.

Methodology:

Implements a deep learning architecture with hierarchical feature fusion of CT and X-ray data and uses Ensemble Monte Carlo Dropout (EMCD) for uncertainty estimation, with evaluation via simulation studies and metrics including precision, accuracy, F-Measure, and ROC curves.