Implementation of convolutional neural network approach for COVID-19 disease detection

Area of Science:

• Artificial Intelligence
• Medical Imaging Analysis
• Deep Learning

Background:

• Accurate and rapid diagnosis of respiratory illnesses like COVID-19 is crucial.
• Chest X-rays are a common diagnostic tool, but manual interpretation can be time-consuming and subjective.
• Existing automated methods often require manual feature extraction, limiting their efficiency.

Purpose of the Study:

• To develop and validate novel Convolutional Neural Network (CNN) architectures for chest X-ray classification.
• To automatically detect COVID-19 from chest X-rays with high accuracy.
• To classify chest X-rays into three categories: COVID-19, normal, and pneumonia.

Main Methods:

• Design and implementation of two distinct CNN architectures.
• Automatic hyperparameter optimization using Grid Search for both models.
• Training and evaluation on a large dataset comprising 1,524 COVID-19, 1,527 pneumonia, and 1,524 normal chest X-ray images.

Main Results:

• The first CNN architecture achieved an average accuracy of 98.92% for COVID-19 detection.
• The second CNN architecture achieved an average accuracy of 98.27% for classifying COVID-19, normal, and pneumonia.
• The proposed models demonstrated effectiveness on large clinical datasets and do not require manual feature extraction.

Conclusions:

• The developed CNN models offer a powerful and robust solution for automated chest X-ray analysis.
• These models represent a significant advancement over existing methods by eliminating the need for diseased tissue extraction.
• This study presents the first CNN-based COVID-19 detection using Grid Search-optimized hyperparameters on the largest available clinical dataset.