An Ensemble Method for Classifying Regional Disease Patterns of Diffuse Interstitial Lung Disease Using HRCT Images...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Computational Pathology

Background:

• Inter-scanner variability poses challenges in analyzing high-resolution computed tomography (HRCT) images for diffuse interstitial lung disease (ILD).
• Accurate differentiation of regional disease patterns is crucial for effective ILD diagnosis and management.
• Standard image analysis methods can be susceptible to variations introduced by different imaging hardware.

Purpose of the Study:

• To evaluate the effectiveness of ensemble classifiers in overcoming inter-scanner variations for ILD pattern differentiation.
• To compare the performance of ensemble classifiers against individual classifiers using both regions of interest (ROIs) and whole lung data.
• To assess the generalizability of the proposed method across intra-scanner, integrated scanner, and inter-scanner experimental settings.

Main Methods:

• Extracted textual and shape features from 600 regions of interest (ROIs) and whole lung areas of 92 HRCT images from GE and Siemens scanners.
• Classified six regional pulmonary disease patterns using expert radiologist annotations.
• Trained and tested individual and ensemble classifiers on ROI and whole lung datasets under three experimental conditions: intra-scanner, integrated scanner, and inter-scanner.

Main Results:

• Ensemble classifiers achieved higher accuracy (89.73%) than individual classifiers (88.38%) in the integrated scanner ROI-based classification (p < 0.001).
• Partial accuracy improvements were observed for ensemble classifiers in intra- and inter-scanner tests.
• Whole lung classification showed higher quantification accuracies with ensemble classifiers (49.57%) compared to individual classifiers (48.19%) using integrated training (p < 0.001).

Conclusions:

• Ensemble classifiers demonstrate superior performance in differentiating ILD patterns from HRCT images, particularly when training data from multiple scanners are integrated.
• The proposed method offers a robust solution to mitigate inter-scanner variability in medical image analysis.
• Ensemble classifiers provide enhanced diagnostic consistency for diffuse interstitial lung disease across diverse imaging equipment.