Multi-class deep learning architecture for COVID-19, tuberculosis, and pneumonia classification using chest X-ray...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonary Medicine

Background:

• Pulmonary diseases like COVID-19, tuberculosis (TB), and pneumonia pose significant health concerns.
• Accurate and timely diagnosis of these conditions is crucial for effective treatment.
• Medical imaging, particularly chest X-rays, is a primary diagnostic tool.

Purpose of the Study:

• To develop and evaluate a deep learning framework for multi-class classification of pulmonary diseases from chest X-ray images.
• To automatically detect COVID-19, TB, pneumonia, and normal conditions.
• To address class imbalance in publicly available datasets.

Main Methods:

• Utilized a convolutional neural network (CNN)-based framework for multi-class classification.
• Applied Synthetic Minority Oversampling Technique (SMOTE) for feature-level data balancing.
• Implemented preprocessing techniques: image normalization, augmentation, and resizing.
• Evaluated multiple deep learning architectures: ResNet-50, EfficientNet, DenseNet, and VGG-19.

Main Results:

• The VGG-19 architecture achieved the highest test accuracy of 97.5%.
• Precision, recall, and F1-scores exceeded 96% across all classes.
• A balanced dataset of 6,000 chest X-ray images was generated using SMOTE.

Conclusions:

• The proposed unified deep learning pipeline effectively integrates data preprocessing, feature extraction, and classification for pulmonary disease detection.
• The VGG-19 model demonstrates significant potential for assisting radiologists in diagnostic decision-making.
• While currently a research framework, the model shows promise for future clinical exploration.