Improving effectiveness of different deep learning-based models for detecting COVID-19 from computed tomography (CT)...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Public Health

Background:

• COVID-19 pandemic poses significant public health challenges.
• Computed tomography (CT) is vital for rapid and accurate COVID-19 diagnosis.
• Deep learning models, like CNNs, require extensive training data for effective COVID-19 detection from CT scans.

Purpose of the Study:

• To address the challenge of limited training data for deep learning models in COVID-19 detection.
• To propose a method for generating synthetic chest CT images using Generative Adversarial Networks (GANs).
• To evaluate the effectiveness of GAN-generated synthetic data in improving deep learning model performance for COVID-19 diagnosis.

Main Methods:

• Utilized segmentation and data augmentation techniques.
• Employed Generative Adversarial Networks (GANs) to synthesize chest CT images from a limited dataset.
• Trained Convolutional Neural Networks (CNNs) with both real and GAN-generated synthetic images.

Main Results:

• A slight increase in accuracy and performance was observed when CNNs were trained on both real and synthetic images on an internal dataset.
• Performance improvements were noted between specific ranges ($$ and $$) on an external dataset.
• The proposed GAN-based method demonstrated promising results for enhancing deep learning model effectiveness.

Conclusions:

• The GAN-based approach effectively generates synthetic CT images to augment limited real datasets.
• The proposed method shows potential to accelerate COVID-19 detection and develop more robust diagnostic systems.
• Further validation is warranted to fully realize the clinical impact of this approach.