COVID-19 Diagnosis in Computerized Tomography (CT) and X-ray Scans Using Capsule Neural Network.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Deep Learning

Background:

• Accurate and rapid COVID-19 diagnosis is crucial for patient management and disease control.
• Traditional deep learning models often struggle with variations in medical image orientation and transformations.
• Capsule Networks (CapsNets) offer inherent robustness to such transformations, making them suitable for medical image analysis.

Purpose of the Study:

• To propose and evaluate a deep-learning-based solution, CapsNetCovid, for COVID-19 diagnosis using capsule neural networks.
• To assess the performance of CapsNetCovid on both standard and augmented COVID-19 CT and X-ray image datasets.
• To compare CapsNetCovid's diagnostic performance against established Convolutional Neural Network (CNN) architectures.

Main Methods:

• Development of CapsNetCovid, a deep learning model based on capsule neural networks.
• Training and evaluation of CapsNetCovid on two COVID-19 datasets comprising CT and X-ray images.
• Performance comparison with CNN, DenseNet121, and ResNet50, particularly on images with random transformations and rotations without data augmentation.

Main Results:

• CapsNetCovid achieved high performance metrics for CT images (accuracy: 99.93%, precision: 99.89%, sensitivity: 100%, F1-score: 99.32%) and X-ray images (accuracy: 94.72%, precision: 93.86%, sensitivity: 92.95%, F1-score: 93.39%).
• CapsNetCovid demonstrated superior performance over CNN, DenseNet121, and ResNet50 when evaluated on CT and X-ray images without data augmentation.
• The model showed robustness in correctly identifying randomly transformed and rotated medical images.

Conclusions:

• CapsNetCovid, utilizing capsule neural networks, presents a highly effective deep learning approach for COVID-19 diagnosis from CT and X-ray images.
• The model's inherent robustness to image transformations offers significant advantages over traditional CNNs, especially in scenarios with limited or no data augmentation.
• This research has the potential to enhance diagnostic accuracy and support clinical decision-making for medical professionals in identifying COVID-19.