Optimized Xception Learning Model and XgBoost Classifier for Detection of Multiclass Chest Disease from X-ray Images

Area of Science:

• Medical Imaging and Artificial Intelligence
• Radiology and Diagnostic Imaging
• Computational Pathology

Background:

• Computed tomography (CT) scans and chest radiography are crucial for diagnosing lung conditions, including coronavirus disease 2019 (COVID-19).
• Existing computer-aided diagnosis (CAD) systems often lack generalizability, require extensive hyper-parameter tuning, and are computationally inefficient for large datasets.
• Deep learning models for medical image analysis face challenges with high computational complexity, memory costs, and background complexities, hindering efficient training.

Purpose of the Study:

• To develop an efficient and accurate artificial intelligence model for classifying chest X-ray images into four categories: normal lungs, lung opacities, COVID-19 infected lungs, and pneumonia.
• To overcome the limitations of existing CAD systems by proposing a novel deep learning architecture.
• To enhance the diagnostic capabilities for radiologists by providing a reliable tool for early detection of lung diseases.

Main Methods:

• Development of a modified Xception (m-Xception) architecture, incorporating depth-separable convolution layers and linear residuals, inspired by the Inception module.
• Implementation of a two-stage transfer learning process for effective model training.
• Utilizing the XgBoost classifier for multi-class recognition of chest X-rays.
• Employing data augmentation techniques to expand a dataset of 1095 images to 48,000, balancing classes for normal lungs, pneumonia, COVID-19, and lung opacities.

Main Results:

• The m-Xception model achieved an average accuracy of 96.5%, F1 score of 96%, recall of 96%, and precision of 96% across various train-test divisions.
• Comparative analysis indicated that the m-Xception method outperformed existing comparable methods in chest X-ray classification.
• The model demonstrated robust performance in identifying and categorizing normal lungs, lung opacities, COVID-19, and pneumonia.

Conclusions:

• The proposed m-Xception model offers a computationally efficient and highly accurate solution for diagnosing multiple lung conditions from chest X-rays.
• This AI-driven approach has the potential to significantly assist radiologists in improving the speed and accuracy of lung disease diagnosis.
• The study highlights the effectiveness of modified deep learning architectures and transfer learning in medical image analysis.