COVID-19 Detection and Diagnosis Model on CT Scans Based on AI Techniques

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonology

Background:

• The COVID-19 pandemic highlighted the need for rapid and accurate diagnostic tools.
• Pulmonary CT scans are crucial for diagnosing respiratory illnesses, but interpretation can be time-consuming.
• Overworked clinicians necessitate automated solutions for efficient medical image analysis.

Purpose of the Study:

• To develop a standardized, deep learning-based reporting system for pulmonary CT scans.
• To classify patients into three categories: COVID-19, viral pneumonia, or healthy.
• To address the challenge of classifying respiratory conditions using medical imaging and AI.

Main Methods:

• Extraction of lung cavities from CT scans followed by data augmentation.
• Development of Convolutional Neural Networks (CNNs) for binary (COVID vs. non-COVID) and three-class classification.
• Utilized transfer learning by integrating a custom linear classifier with pre-trained models like MobileNetV1, ResNet50, DenseNet201, VGG16, and VGG19.

Main Results:

• MobileNetV1 achieved superior detection accuracy for pulmonary anomalies using the CAM technique.
• The model demonstrated high performance with an accuracy of 0.9744, precision of 0.9758, recall of 0.9742, AUC of 0.9902, and F1 score of 0.9750.
• Achieved excellent classification results on 1161 samples per class for COVID-19, viral pneumonia, and healthy individuals.

Conclusions:

• Deep learning, particularly with CNNs and transfer learning, offers a promising solution for automated pulmonary CT scan analysis.
• MobileNetV1 shows significant potential for accurate COVID-19 and viral pneumonia detection, even when not widely used in existing literature.
• The developed system can aid in efficient and accurate diagnosis, potentially alleviating clinician workload.