Genetic-based adaptive momentum estimation for predicting mortality risk factors for COVID-19 patients using deep...

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Radiology

Background:

• Early prediction of mortality risk factors for coronavirus disease (COVID-19) is essential for timely intensive care.
• Identifying severe cases before critical illness allows for proactive medical intervention.

Purpose of the Study:

• To develop an optimized Convolutional Neural Network (CNN) model for predicting COVID-19 mortality risk factors.
• To integrate both clinical variables and Computed Tomography (CT) scan data for enhanced prediction accuracy.

Main Methods:

• Utilized an optimized CNN model with feature extraction from clinical data and CT scans.
• Employed a novel Genetic-Based Adaptive Momentum Estimation (GB-ADAM) algorithm to optimize CNN hyperparameters, incorporating a Genetic Algorithm (GA).
• Validated the model on three large, diverse cohorts (New York, Mexico, Wuhan).

Main Results:

• Identified key mortality risk factors: CD T Lymphocyte Count, D-dimer levels (>1 Ug/ml), lactate dehydrogenase (LDH), C-reactive protein (CRP), hypertension, and diabetes.
• Frequent COVID-19 CT scan findings included ground-glass opacity (GGO), crazy-paving pattern, consolidations, and lobe count.
• The proposed model demonstrated encouraging performance compared to existing prediction models.

Conclusions:

• Early identification of these significant risk factors can significantly aid clinicians in providing immediate and effective care to COVID-19 patients.
• The developed CNN model offers a promising tool for predicting COVID-19 mortality, integrating multi-modal data for improved accuracy.