CapsNet-COVID19: Lung CT image classification method based on CapsNet model

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonology

Background:

• The COVID-19 pandemic presents diagnostic challenges, including RT-PCR errors and reagent shortages.
• Accurate and efficient diagnostic methods are crucial for managing the global health crisis.
• Current diagnostic limitations necessitate supplementary tools for COVID-19 detection.

Purpose of the Study:

• To develop and evaluate a deep learning model for classifying COVID-19, pneumonia, and normal lung CT images.
• To address limitations in traditional COVID-19 testing methods.
• To provide a supplementary diagnostic tool for medical image analysis.

Main Methods:

• Utilized a two-level deep network model for lung CT image classification.
• Employed convolutional neural networks for lung region localization.
• Applied capsule networks for classification and prediction of segmented images.

Main Results:

• Achieved 84.291% accuracy on the test set and 100% on the training set.
• Demonstrated suitability for medical images with complex backgrounds and noise.
• Showcased effectiveness in classifying blurred boundaries and low-recognition rate images.

Conclusions:

• The proposed deep network model is a valuable tool for classifying lung CT images.
• This method shows promise as an accurate and efficient supplement to COVID-19 diagnosis.
• The approach is beneficial for monitoring and controlling the spread of COVID-19 infections.