Computer-aided classification of interstitial lung diseases via MDCT: 3D adaptive multiple feature method (3D AMFM)

Area of Science:

• Radiology and Medical Imaging
• Computer-Aided Diagnosis
• Pulmonary Medicine

Background:

• Multidetector row CT (MDCT) enables quantitative analysis of lung image data.
• Computer-aided detection (CAD) tools can aid in early detection of pulmonary pathology.
• The adaptive multiple feature method (AMFM) is a developed CAD tool for lung disease detection.

Purpose of the Study:

• To further develop the AMFM CAD tool for interstitial lung disease detection using MDCT data.
• To evaluate the efficacy of volumetric features in differentiating lung pathologies.
• To compare classification performance of Support Vector Machine (SVM) and Bayesian methods.

Main Methods:

• MDCT scans were acquired from 20 volunteers across four cohorts: normal never-smokers, normal smokers, emphysema, and interstitial lung disease.
• A senior radiologist and pulmonologist marked 1,184 volumes of interest (VOIs) for specific lung conditions (emphysema, ground-glass, honeycombing, normal).
• 24 volumetric features (statistical, histogram, fractal) were calculated for each VOI, and classified using SVM and Bayesian methods with 10-fold cross-validation.

Main Results:

• High sensitivity and specificity were achieved for emphysema, ground-glass, and honeycombing patterns using both classifiers.
• Sensitivity for normal non-smokers and smokers varied between classifiers (90%/73% and 75%/82% respectively).
• Specificity for normal non-smokers and smokers was generally high (90-96%) across both classifiers.

Conclusions:

• Volumetric features (statistical, histogram, fractal) are effective for differentiating emphysema and interstitial lung diseases on MDCT.
• SVM and Bayesian classifiers demonstrate comparable performance in characterizing interstitial lung diseases.
• The developed AMFM tool shows promise for quantitative analysis and early detection of pulmonary pathologies.