A novel comparative study for detection of Covid-19 on CT lung images using texture analysis, machine learning, and...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Infectious Diseases

Background:

• The COVID-19 pandemic caused severe pneumonia, necessitating rapid diagnostic tools.
• Lung CT imaging is crucial for visualizing COVID-19's pulmonary effects.
• Early and accurate diagnosis is vital for effective patient management and disease control.

Purpose of the Study:

• To evaluate the efficacy of deep learning and machine learning models for automated COVID-19 classification using lung CT images.
• To compare the performance of Convolutional Neural Network (CNN) architectures against traditional machine learning methods.
• To assess the impact of data augmentation and cross-validation strategies on classification accuracy.

Main Methods:

• A dataset of 1,396 lung CT images (386 COVID-19, 1,010 non-COVID-19) was utilized.
• A custom 23-layer CNN was designed, alongside Alexnet and Mobilenetv2 architectures.
• k-Nearest Neighbors (k-NN) and Support Vector Machine (SVM) were employed for comparative analysis.
• Data augmentation and 2-fold/10-fold cross-validation were implemented to enhance model robustness.

Main Results:

• Deep learning models, especially CNNs, demonstrated superior performance in classifying COVID-19 from CT images.
• The highest accuracy achieved was 0.9599 with 10-fold cross-validation using a CNN approach.
• Sensitivity, specificity, F-1 score, and AUC values consistently indicated high classification success for deep learning methods.

Conclusions:

• Automated classification of lung CT images using deep learning, particularly CNNs, is a highly effective method for early COVID-19 diagnosis.
• The study validates the potential of AI-driven tools to aid radiologists in rapid and accurate COVID-19 detection.
• Optimized training strategies, including data augmentation and cross-validation, significantly improve diagnostic performance.