DIAG a Diagnostic Web Application Based on Lung CT Scan Images and Deep Learning

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Science

Background:

• The COVID-19 pandemic necessitates efficient diagnostic tools.
• Lung CT scans are crucial for diagnosis but can overwhelm radiologists.
• Automated image interpretation is needed to support clinical workflows.

Purpose of the Study:

• To develop and evaluate deep learning models for classifying COVID-19 positive patients from healthy individuals using lung CT scans.
• To assess the generalizability of the developed models across different datasets.
• To propose a computer-aided diagnosis (CAD) web application for COVID-19 detection.

Main Methods:

• Collected and utilized four publicly available datasets of lung CT scans.
• Trained and tested convolutional neural networks (CNNs) on various data distributions.
• Employed Grad-CAM and Fast-CAM visualization techniques to interpret model predictions.
• Developed a computer-aided diagnosis web application.

Main Results:

• Achieved over 92% accuracy in classifying COVID-19 positive and healthy patients across two different data distributions.
• Demonstrated the generalizability of the CNN models.
• Successfully visualized model decision-making processes using Grad-CAM and Fast-CAM.

Conclusions:

• Deep learning models show high accuracy and generalizability for COVID-19 detection from lung CT scans.
• The proposed CAD web application can aid in rapid patient management.
• The developed DL tool has the potential for integration into clinical settings to assist in COVID-19 diagnosis.