Stacked-autoencoder-based model for COVID-19 diagnosis on CT images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Science

Background:

• Computed tomography (CT) imaging is crucial for diagnosing COVID-19.
• Deep learning models show promise for CT-based COVID-19 detection.
• Limited public datasets due to privacy concerns hinder deep learning model performance.

Purpose of the Study:

• To develop an improved deep learning model for COVID-19 detection using CT images.
• To address the challenge of limited data availability in COVID-19 CT imaging research.
• To enhance the precision and recall rates of infectious disease detection models.

Main Methods:

• Proposed a stacked autoencoder detector model for feature extraction from CT images.
• Cascaded four autoencoders, connected to a dense layer and softmax classifier.
• Introduced a novel classification loss function by superimposing reconstruction loss.

Main Results:

• Achieved high performance on a small COVID-19 CT image dataset.
• Reported average accuracy of 94.7%, precision of 96.54%, recall of 94.1%, and F1-score of 94.8%.
• Demonstrated effective discrimination of COVID-19 positive CT images.

Conclusions:

• The proposed stacked autoencoder model significantly improves COVID-19 detection accuracy.
• The model's performance on small datasets suggests its utility in real-world clinical settings.
• This approach can assist radiologists in diagnosing suspected COVID-19 cases.