Computer-aided diagnosis of lung nodule using gradient tree boosting and Bayesian optimization.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Computer-aided diagnosis (CADx) systems are crucial for lung nodule classification.
• Evaluating conventional CADx systems, machine learning algorithms, and parameter optimization is essential.

Purpose of the Study:

• To assess the utility of a conventional CADx system for lung nodule classification.
• To compare Support Vector Machine (SVM) and Gradient Tree Boosting (XGBoost) algorithms.
• To evaluate the effectiveness of Bayesian optimization versus random search for parameter tuning.

Main Methods:

• Utilized CT image data from 99 lung nodules (62 cancerous, 37 benign) from public databases.
• Employed a variant of the local binary pattern for feature vector calculation.
• Trained SVM and XGBoost models, optimizing parameters using Tree Parzen Estimator (TPE) Bayesian optimization and random search, with leave-one-out cross-validation.

Main Results:

• XGBoost achieved a superior average Area Under the Curve (AUC) of 0.896 compared to SVM's 0.850, both optimized with TPE.
• Bayesian optimization (TPE) was more efficient than random search, requiring fewer trials to find optimal parameters.
• The CADx system's diagnostic accuracy, measured by AUC, was comparable to that of board-certified radiologists (0.898 and 0.822).

Conclusions:

• The developed CADx system demonstrates high diagnostic accuracy for lung nodule classification.
• XGBoost, optimized via Bayesian methods, outperforms SVM for this task.
• The CADx system's performance is on par with experienced radiologists, highlighting its clinical potential.