Convolutional neural network based CT scan classification method for COVID-19 test validation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Computed Tomography (CT) is crucial for diagnosing COVID-19 due to RT-PCR limitations.
• Limited COVID-19 CT datasets pose challenges for developing accurate diagnostic models.
• Overfitting is a significant risk when training models on small medical image datasets.

Purpose of the Study:

• To address the scarcity of COVID-19 CT datasets by employing data augmentation techniques.
• To develop an improved deep learning model for accurate COVID-19 detection using CT images.
• To enhance the diagnostic capabilities of radiologists through advanced AI in medical imaging.

Main Methods:

• Utilized Conditional Generative Adversarial Networks (CGANs) for augmenting limited COVID-19 CT datasets.
• Developed a novel BUF-Net architecture integrating a U-Net network with BIN residual blocks.
• Employed multi-layer perception for classification prediction and Grad-CAM for result visualization.

Main Results:

• The proposed BUF-Net model achieved a high accuracy rate of 93% in COVID-19 detection.
• Data augmentation using CGANs effectively increased the sample size of the CT dataset, reducing overfitting risk.
• Grad-CAM visualization confirmed the critical role of CT features in the diagnostic process.

Conclusions:

• The developed deep learning techniques, including CGANs and the BUF-Net model, offer a promising approach for COVID-19 detection from CT images.
• AI-powered medical image analysis can significantly aid radiologists in achieving more effective and timely diagnoses.
• This research highlights the potential of advanced AI in overcoming data scarcity challenges in medical diagnostics.