Improving Tree-Based Lung Disease Classification from Chest X-Ray Images Using Deep Feature Representations.

Area of Science:

• Medical Imaging and Diagnostics
• Artificial Intelligence in Healthcare
• Computational Pathology

Background:

• Healthcare systems require scalable, accurate, and affordable diagnostic tools.
• Chest X-rays are cost-effective for lung disease screening.
• Current deep learning models are often computationally intensive and lack interpretability, hindering clinical adoption.

Purpose of the Study:

• To develop a hybrid Convolutional Neural Network (CNN)-tree framework for automated multi-class lung disease classification from chest X-ray images.
• To address challenges of computational intensity and interpretability in deep learning for diagnostics.
• To target specific lung conditions including COVID-19, pneumonia, tuberculosis, and lung cancer.

Main Methods:

• A unified five-class dataset was created by merging four public chest X-ray datasets.
• A ResNet-18 model was fine-tuned for deep feature extraction.
• Principal Component Analysis (PCA) and Synthetic Minority Over-sampling Technique (SMOTE) were used for dimensionality reduction and class imbalance.
• Interpretable tree-based classifiers (Decision Tree, Random Forest, XGBoost) were trained on the processed features.

Main Results:

• The hybrid framework achieved high performance, with weighted F1-scores ranging from 0.977 to 0.982 across tree-based classifiers.
• Significant reduction in inter-class confusion was observed.
• The model demonstrated low per-sample inference latency, supporting energy-efficient deployment.

Conclusions:

• Combining deep feature learning with interpretable tree-based models offers a practical and reliable solution for sustainable chest X-ray screening.
• The proposed framework enhances the accuracy and interpretability of automated lung disease classification.
• This approach is suitable for high-throughput, resource-constrained clinical environments.