A novel deep fusion strategy for COVID-19 prediction using multimodality approach

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Respiratory Medicine

Background:

• The novel coronavirus (COVID-19) poses a significant global health threat.
• Accurate and timely symptom determination is crucial for managing the pandemic.
• Existing diagnostic methods face challenges with the complexity and variability of COVID-19 symptoms.

Purpose of the Study:

• To introduce a Deep Learning-assisted Multi-modal Data Analysis (DMDA) approach for COVID-19 symptom detection.
• To leverage acoustic and image-based data for enhanced diagnostic capabilities.
• To validate the efficacy of a Dynamic Fusion Strategy (DFS) for confirming individual health status.

Main Methods:

• Development of a DMDA approach integrating deep learning models.
• Utilizing acoustic and image-based data as input modalities.
• Implementation of a Dynamic Fusion Strategy (DFS) for final health status determination.

Main Results:

• The DMDA approach achieved high accuracy on individual data modalities: 96.88% for acoustic and 98.76% for image-based data.
• The proposed Dynamic Fusion Strategy (DFS) demonstrated an overall symptom determination accuracy of 98.72%.
• The system maintained high reliability, achieving 95.64% accuracy even when one data modality was absent during testing.

Conclusions:

• The proposed DMDA approach, coupled with DFS, offers a robust and accurate method for COVID-19 symptom determination.
• Multi-modal data analysis significantly enhances diagnostic performance in the context of complex infectious diseases.
• The developed system shows promise for reliable, AI-assisted decision-making in public health surveillance.