Transfer learning-based approach for detecting COVID-19 ailment in lung CT scan

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Accurate and efficient identification of COVID-19 is crucial for public health management.
• Lung CT scans are a valuable tool for diagnosing COVID-19, but image quality can impact diagnostic accuracy.

Purpose of the Study:

• To develop and evaluate deep learning models for COVID-19 identification using lung CT scan images.
• To enhance the efficiency and accuracy of lung CT scan analysis for COVID-19 detection through super-resolution techniques.

Main Methods:

• Application of a super-residual dense neural network for lung CT scan enhancement.
• Utilized benchmark datasets: SARS-COV-2 CT-Scan and Covid-CT Scan.
• Employed pre-trained deep learning models (XceptionNet, MobileNet, InceptionV3, DenseNet, ResNet50, VGG16) for classification.

Main Results:

• Super-resolution pre-processing using a residual dense neural network improved accuracy, F1 score, precision, and recall.
• The MobileNet model achieved high precision, reaching 94.12% on the Covid-CT Scan dataset and 100% on the SARS-COV-2 CT-Scan dataset.

Conclusions:

• Deep learning models, particularly when combined with super-resolution pre-processing, show significant promise for accurate COVID-19 detection from lung CT scans.
• The enhanced CT scan analysis improves diagnostic performance, aiding in more reliable identification of the virus.