An Expert System for COVID-19 Infection Tracking in Lungs Using Image Processing and Deep Learning Techniques.

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Accurate detection of respiratory diseases like COVID-19 and pneumonia from X-ray images is crucial for timely treatment.
• Raw X-ray images can present challenges in feature extraction and segmentation, potentially impacting diagnostic accuracy.
• Existing machine and deep learning models may benefit from improved image quality and lung segmentation.

Purpose of the Study:

• To develop and evaluate an efficient preprocessing and classification technique for respiratory disease detection using X-ray lung images.
• To enhance the accuracy of classifying normal, COVID-19, and pneumonia cases through optimized image preprocessing and lung segmentation.
• To compare the performance of various deep learning architectures, including a proposed deep neural network, for respiratory disease classification.

Main Methods:

• Algorithms including Histogram of Oriented Gradients (HOG), Haar Transform (Haar), and Local Binary Pattern (LBP) were applied for feature extraction and lung segmentation.
• Intersection over Union (IoU) scores were used to validate the accuracy of lung segmentation across different algorithms.
• Deep learning models such as VGGNet, AlexNet, Resnet, and a proposed deep neural network were implemented for image classification.

Main Results:

• Image preprocessing significantly improved the accuracy of lung segmentation, as indicated by higher Intersection over Union scores.
• Preprocessed X-ray lung images demonstrated superior classification accuracy for all three classes (normal, COVID-19, pneumonia) compared to raw images.
• The proposed deep neural network architecture achieved higher classification accuracy than VGGNet, AlexNet, and Resnet for respiratory disease detection.

Conclusions:

• The proposed image preprocessing algorithms effectively enhance the quality of X-ray lung images for improved feature extraction and segmentation.
• The integration of advanced preprocessing techniques with deep learning models offers a promising approach for accurate and efficient respiratory disease detection.
• The developed deep neural network shows significant potential for clinical application in diagnosing respiratory conditions from X-ray imagery.