COVID-19 infection segmentation using hybrid deep learning and image processing techniques.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• The COVID-19 pandemic necessitates rapid and accurate diagnostic tools.
• Traditional medical imaging analysis for COVID-19 faces challenges in accuracy and segmentation.
• Deep learning offers potential for improving diagnostic capabilities in medical imaging.

Purpose of the Study:

• To develop and evaluate a novel deep learning approach for enhanced COVID-19 detection in lung images.
• To improve the accuracy and efficiency of identifying COVID-19 infections using computed tomography (CT) scans.
• To leverage image processing techniques and U-Net architecture for precise infection segmentation.

Main Methods:

• Preprocessing CT images using thresholding and resizing to 128x128 pixels.
• Applying a density heat map for coloring, followed by RGB channel separation.
• Utilizing three U-Net models for independent channel segmentation and combining results via convolution.

Main Results:

• The proposed method achieved high performance metrics: 99.71% accuracy, 0.83 sensitivity, 0.87 precision, and 0.85 dice coefficient.
• Coloring CT images and processing RGB channels improved U-Net segmentation effectiveness.
• The approach demonstrated accurate detection on smaller 128x128 images compared to methods using larger 512x512 images.

Conclusions:

• The novel deep learning method effectively enhances COVID-19 detection through image processing and U-Net segmentation.
• Processing RGB channels of colored CT images significantly boosts detection accuracy and segmentation power.
• This technique offers a rapid and highly accurate solution for COVID-19 diagnosis using medical imaging.