Lung disease stratification using amalgamation of Riesz and Gabor transforms in machine learning framework

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Pulmonary Medicine

Background:

• Lung disease risk stratification is crucial for diagnosis and treatment planning.
• Manual methods face challenges due to large data, observer variability, and limited feature integration.
• Existing machine learning models often lack comprehensive feature amalgamation.

Purpose of the Study:

• To develop and evaluate a two-stage Computer-Aided Diagnosis (CADx) system for lung disease risk stratification.
• To address limitations of manual stratification and conventional machine learning approaches.
• To improve accuracy and robustness in identifying lung disease risks.

Main Methods:

• A semi-automated lung delineation subsystem (LDS) using entropy-based region extraction for CT slices.
• Morphology-based lung tissue characterization using amalgamated directional (Riesz, Gabor) and texture features.
• K-fold cross-validation (K=2, 3, 5, 10) on a database of 96 patients (15 normal, 81 diseased) across five HRCT levels.

Main Results:

• The CADx system achieved a lung disease risk stratification accuracy of 99.53%.
• This represents a 2% improvement over conventional non-amalgamation machine learning systems.
• The system demonstrated high robustness with a reliability index of 0.99 and risk stratification accuracy deviation <5%.

Conclusions:

• The proposed two-stage CADx system effectively enhances lung disease risk stratification accuracy.
• Amalgamation of diverse features significantly improves machine learning performance in lung imaging.
• This approach offers a robust and accurate alternative to manual methods, outperforming existing prominent studies.