Performance testing of several classifiers for differentiating obstructive lung diseases based on texture analysis at...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Pulmonary Medicine

Background:

• Machine classifiers automate quantitative analysis in medical imaging, reducing reader variability.
• Accurate classification schemes are crucial for improving diagnostic performance based on dataset characteristics.
• Obstructive lung diseases, including emphysema and bronchiolitis obliterans, require precise differentiation.

Purpose of the Study:

• To investigate the performance of various machine classifiers in differentiating obstructive lung diseases.
• To evaluate texture analysis on different region of interest (ROI) sizes for disease classification.
• To compare the accuracy of Naïve Bayesian, Bayesian, Artificial Neural Network (ANN), and Support Vector Machine (SVM) classifiers.

Main Methods:

• Utilized 265 high-resolution computerized tomography (HRCT) images from 92 subjects.
• Radiologists identified ROIs representing severe and mild centrilobular emphysema, bronchiolitis obliterans, and normal lung.
• Implemented and compared four classifiers (Naïve Bayesian, Bayesian, ANN, SVM) using 5-fold cross-validation repeated 20 times.

Main Results:

• Support Vector Machine (SVM) demonstrated the highest overall accuracy, particularly with 32x32 and 64x64 ROI sizes (p<0.05).
• No significant accuracy difference was found between Bayesian and ANN classifiers (p<0.05).
• Naïve Bayesian classifier performed significantly worse than the other tested methods (p<0.05).

Conclusions:

• SVM achieved the best performance for classifying obstructive lung diseases using texture analysis on HRCT images.
• The choice of machine classifier significantly impacts diagnostic accuracy in pulmonary imaging.
• Texture analysis combined with SVM offers a promising automated approach for obstructive lung disease differentiation.