COVID-19 Pneumonia Diagnosis Using a Simple 2D Deep Learning Framework With a Single Chest CT Image: Model...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Computed tomography (CT) is vital for detecting early COVID-19 pneumonia.
• Physician workload during the COVID-19 pandemic necessitates AI support for diagnostics.

Purpose of the Study:

• To develop an AI tool for rapid COVID-19 pneumonia diagnosis using CT images.
• To differentiate COVID-19 pneumonia from other conditions.

Main Methods:

• A 2D deep learning framework, FCONet, was created using transfer learning with pretrained models (VGG16, ResNet-50, Inception-v3, Xception).
• Trained and tested on 3993 CT images from multiple databases, with an 8:2 train-test split.
• Performance evaluated on internal and external datasets, including low-quality CT images.

Main Results:

• The ResNet-50 based FCONet model demonstrated superior performance with 99.58% sensitivity, 100.00% specificity, and 99.87% accuracy on the internal test set.
• On an external dataset of low-quality CT images, the ResNet-50 model achieved the highest detection accuracy at 96.97%.

Conclusions:

• FCONet, utilizing a single CT image, offers excellent diagnostic performance for COVID-19 pneumonia.
• The ResNet-50 backbone provides the optimal FCONet model for COVID-19 pneumonia detection.