Leveraging deep learning for COVID-19 diagnosis through chest imaging

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Infectious Disease Diagnostics

Background:

• The COVID-19 pandemic necessitates accurate and timely diagnosis.
• Reverse transcription-polymerase chain reaction (RT-PCR) is the gold standard but has limitations, including false negatives.
• Medical imaging, particularly chest imaging, is increasingly considered to aid COVID-19 diagnosis.

Purpose of the Study:

• To evaluate the efficacy of computer vision models in detecting COVID-19 from chest radiographic images.
• To compare the performance of different deep learning models for COVID-19 classification.
• To assess the potential of AI-assisted medical imaging as a supplementary diagnostic tool.

Main Methods:

• Utilized chest computed tomography (CT) scans and chest X-ray images for analysis.
• Compared performance of ResNet-50, EfficientNetB0, VGG-16, and a custom convolutional neural network (CNN).
• Focused on classifying images to detect the presence of the virus.

Main Results:

• ResNet-50 model demonstrated high accuracy on both CT scans (98.9%) and X-rays (98.7%).
• The study identified ResNet-50 as a promising model for COVID-19 detection in radiographic images.
• AI models showed significant potential in aiding the diagnostic process.

Conclusions:

• Chest imaging combined with AI, specifically ResNet-50, shows high accuracy for COVID-19 detection.
• These AI-driven methods can serve as a valuable aid to physicians, complementing RT-PCR testing.
• Further research into these prospective methods is warranted for clinical application.