Improving CNN predictive accuracy in COVID-19 health analytics.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Epidemiology

Background:

• The COVID-19 pandemic necessitates advanced predictive models for healthcare.
• Convolutional Neural Networks (CNNs) offer potential for analyzing complex health data.

Purpose of the Study:

• To evaluate CNNs for COVID-19 infection identification via chest X-rays.
• To assess CNNs for predicting COVID-19 severity using clinical and electronic health records.

Main Methods:

• Utilized a multidimensional dataset including demographic, clinical, and radiological information.
• Applied CNN architectures to forecast patient prognoses and identify infections.
• Evaluated model performance using accuracy, precision, F1-score, and AUC-ROC metrics.

Main Results:

• CNN model achieved 97.2% accuracy, 96.8% precision, 97.2% F1-score, and 0.987 AUC-ROC.
• Identified challenges: limited/imbalanced data, model interpretability, overfitting, and data heterogeneity.
• Proposed strategies like data augmentation and transfer learning improved performance.

Conclusions:

• CNNs are effective tools for COVID-19 detection and prognosis.
• Continuous refinement and domain expertise are crucial for CNN-based healthcare analytics.
• Findings guide data-driven approaches in managing COVID-19 outcomes.