Densely connected convolutional networks-based COVID-19 screening model.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Infectious Disease Diagnostics

Background:

• Real-time polymerase chain reaction (RT-PCR) is the standard for COVID-19 detection but is time-consuming, costly, and prone to false negatives.
• Deep learning models offer potential for early-stage classification of suspected COVID-19 cases, addressing RT-PCR's limitations.
• Chest CT scans reveal unique patterns in COVID-19 patients, but manual analysis is labor-intensive.

Purpose of the Study:

• To develop and evaluate an automated COVID-19 screening model using ensemble deep transfer learning on chest CT scans.
• To improve the speed, accuracy, and sensitivity of COVID-19 diagnosis compared to traditional methods.
• To differentiate COVID-19 positive cases from tuberculosis, pneumonia, and healthy subjects.

Main Methods:

• An ensemble model was created by combining Densely connected convolutional networks (DCCNs), ResNet152V2, and VGG16.
• The model was trained and validated using chest CT scan data from various patient groups.
• Performance was evaluated based on accuracy, f-measure, area under the curve, sensitivity, and specificity.

Main Results:

• The proposed ensemble model demonstrated superior performance over individual models and competitive methods.
• The model achieved high accuracy, f-measure, area under the curve, sensitivity, and specificity in classifying COVID-19.
• The automated approach offers a faster and potentially more reliable alternative to RT-PCR for COVID-19 screening.

Conclusions:

• Ensemble deep transfer learning models applied to chest CT scans provide an effective automated solution for COVID-19 screening.
• This approach addresses the limitations of RT-PCR, including cost, time, and false-negative rates.
• The developed model shows promise for early and accurate detection of COVID-19, aiding in clinical decision-making.