Deep Transfer Learning Based Classification Model for COVID-19 Disease.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Infectious Diseases

Background:

• COVID-19 infection rates are rising globally, straining diagnostic resources.
• Chest Computed Tomography (CT) imaging shows characteristic bilateral changes in COVID-19 patients.
• Classifying COVID-19 from CT scans is challenging due to the ill-posed nature of bilateral change prediction.

Purpose of the Study:

• To develop an efficient deep transfer learning model for classifying COVID-19 patients using chest CT images.
• To address the challenges of noisy and imbalanced datasets in COVID-19 classification.

Main Methods:

• A deep transfer learning technique was employed for COVID-19 patient classification.
• A top-2 smooth loss function with cost-sensitive attributes was utilized to manage dataset noise and imbalance.
• The model was trained and evaluated on chest CT image data.

Main Results:

• The proposed deep transfer learning model demonstrated efficient classification performance.
• Experimental results indicate superior efficiency compared to traditional supervised learning models.
• The model effectively handles noisy and imbalanced COVID-19 datasets.

Conclusions:

• Deep transfer learning offers a promising approach for COVID-19 classification from chest CT scans.
• The developed model provides an efficient alternative for COVID-19 diagnosis, especially when testing kits are limited.
• The use of specialized loss functions enhances model robustness for real-world medical datasets.