COVID-19 Semantic Pneumonia Segmentation and Classification Using Artificial Intelligence

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Science

Background:

• Coronavirus 2019 (COVID-19) is a global pandemic with significant mortality.
• Accurate and efficient diagnostic tools are crucial for managing the pandemic.
• Deep learning offers potential for automated analysis of medical images.

Purpose of the Study:

• To develop and evaluate an efficient deep semantic segmentation network for COVID-19 detection using chest X-rays.
• To optimize a custom convolutional neural network model for improved accuracy and efficiency.
• To assess the performance of image enhancement and data augmentation techniques in COVID-19 diagnosis.

Main Methods:

• Utilized dynamic adaptive histogram equalization for image enhancement.
• Applied data augmentation techniques to increase dataset variability.
• Developed a custom convolutional neural network (CNN) by integrating and refining pretrained ImageNet models.
• Compared multiple model variations to select the most efficient and accurate configuration.

Main Results:

• The proposed model achieved a high accuracy of 99.6% for COVID-19 detection.
• The model demonstrated a strong performance with an area under the curve (AUC) of 0.996.
• Optimized model complexity and memory efficiency through iterative trimming of well-performing components.

Conclusions:

• The developed customized smart convolutional neural network is highly effective for COVID-19 detection from chest X-rays.
• The integration of image enhancement and data augmentation significantly contributes to diagnostic accuracy.
• This deep learning approach shows promise for rapid and reliable screening of COVID-19.