Early Diagnosis of COVID-19 Images Using Optimal CNN Hyperparameters.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Science

Background:

• Coronavirus disease (COVID-19) presents significant global health challenges.
• Current COVID-19 diagnostic test sensitivity is limited by specimen processing issues.
• Optimizing convolutional neural network (CNN) hyperparameters is crucial for enhancing diagnostic performance.

Purpose of the Study:

• To propose an optimization strategy for CNN hyperparameters, specifically learning rate and momentum, using grid search.
• To improve the performance and accuracy of COVID-19 detection through hyperparameter optimization.
• To evaluate the effectiveness of optimized CNN architectures on diverse COVID-19 radiography datasets.

Main Methods:

• Implemented grid search to optimize learning rate and momentum for CNN hyperparameters.
• Utilized three CNN architectures: GoogleNet, VGG16, and ResNet.
• Tested models on two COVID-19 radiography datasets: Kaggle (X-ray) and China national center for bio-information (CT).

Main Results:

• Optimized CNN hyperparameters significantly improved disease classification accuracy.
• The proposed optimization technique outperformed previous methods across various metrics including accuracy, sensitivity, and specificity.
• Optimized ResNet achieved high classification accuracy: 98.98% for X-ray and 98.78% for CT images at epoch 25.

Conclusions:

• Hyperparameter optimization using grid search is an effective strategy for enhancing CNN performance in COVID-19 detection.
• Optimized CNN models demonstrate superior diagnostic capabilities compared to non-optimized models.
• The study highlights the potential of AI-driven approaches for accurate and sensitive COVID-19 diagnosis from radiographic images.