COVIDXception-Net: A Bayesian Optimization-Based Deep Learning Approach to Diagnose COVID-19 from X-Ray Images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Chest X-rays are crucial for COVID-19 diagnosis.
• Limited access to radiological services globally hinders widespread diagnosis.
• Scarcity of COVID-19 data negatively impacts deep learning model performance.

Purpose of the Study:

• To develop an automated deep learning system (COVIDXception-Net) for COVID-19 diagnosis using chest X-rays.
• To overcome the challenge of limited COVID-19 data in deep learning models.
• To enhance diagnostic accuracy in resource-limited settings.

Main Methods:

• Development of COVIDXception-Net, a deep learning model for chest X-ray analysis.
• Implementation of a weighted-loss function to prioritize COVID-19 cases during training.
• Utilization of Bayesian Optimization for optimal model architecture selection.
• Extensive experimentation on four publicly available COVID-19 datasets.

Main Results:

• COVIDXception-Net achieved high diagnostic performance: 0.94 accuracy, 0.95 precision, 0.94 recall, 0.997 specificity, 0.94 F1-score, and 0.992 MCC.
• The model outperformed VGG16, MobileNetV2, and InceptionV3 architectures.
• Ablation studies confirmed the effectiveness of weighted loss and Bayesian optimization, showing performance degradation with standard methods.

Conclusions:

• COVIDXception-Net offers a robust and accurate automated solution for COVID-19 detection from chest X-rays.
• The proposed methods effectively address data scarcity challenges in medical AI.
• This system has the potential to improve COVID-19 diagnosis accessibility, especially where radiological services are limited.