Deep Transfer Learning for COVID-19 Detection and Lesion Recognition Using Chest CT Images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• The COVID-19 pandemic necessitates rapid and accurate diagnostic tools.
• Chest CT scans are crucial for COVID-19 diagnosis but manual review is time-consuming.
• Automated analysis of CT scans can significantly improve diagnostic efficiency.

Purpose of the Study:

• To develop and evaluate a novel deep learning-based method for automated COVID-19 detection using chest CT scans.
• To enhance the performance of deep learning models for COVID-19 diagnosis through architectural improvements.
• To provide a tool that assists radiologists in faster and more accurate COVID-19 diagnosis.

Main Methods:

• Utilized state-of-the-art deep convolutional neural network (CNN) architectures.
• Incorporated a 2D global max pooling (globalMaxPool2D) layer to enhance model performance.
• Compared proposed models against existing CNN and Vision Transformer (ViT) models.

Main Results:

• The proposed models outperformed previous methods in COVID-19 detection from CT scans.
• The best model achieved an Area Under the Curve (AUC) of 0.9744 and 94.12% accuracy.
• The 2D global max pooling layer improved accuracy by approximately 1%.

Conclusions:

• The developed deep learning models offer a fast and accurate solution for COVID-19 diagnosis via chest CT.
• A heatmap method was introduced to visualize lesion areas, aiding radiologist interpretation.
• A freely accessible online software tool was developed to support automated COVID-19 detection.