A semi-supervised learning approach for COVID-19 detection from chest CT scans

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Infectious Disease Diagnostics

Background:

• Early screening of COVID-19 patients is crucial for disease control.
• Chest Computed Tomography (CT) offers rapid diagnostic information.
• High workload for radiologists in interpreting CT scans necessitates automated solutions.

Purpose of the Study:

• To develop a high-precision AI algorithm for intelligent COVID-19 diagnosis from chest CT scans.
• To address challenges with limited labeled data using a semi-supervised learning approach.
• To enhance model generalization and reduce overfitting in COVID-19 detection.

Main Methods:

• A semi-supervised learning framework incorporating MixMatch data augmentation and a novel regularization technique.
• Development of an attention-based convolutional neural network for multi-scale feature extraction from CT scans.
• Rigorous evaluation on an independent chest CT dataset for COVID-19 diagnosis.

Main Results:

• The algorithm achieved an Area Under the ROC Curve (AUC) of 0.932.
• Demonstrated high diagnostic performance with 90.1% accuracy, 91.4% sensitivity, 88.9% specificity, and 89.9% F1-score.
• The model accurately differentiates between COVID-19 positive and negative CT scans.

Conclusions:

• The proposed AI algorithm provides accurate and rapid diagnosis of COVID-19 from chest CTs.
• This tool can significantly assist clinicians in early outbreak detection and patient management.
• The semi-supervised learning approach effectively handles limited labeled medical data for AI model training.