Lung CT image based automatic technique for COPD GOLD stage assessment

Image based analysis of the lung air can be used for lung function assessment and effective diagnosis of lung diseases including chronic obstructive pulmonary disease (COPD). A novel expert system technique is proposed to accurately assess COPD severity characterized by its stage through processing the patients thoracic CT images. The technique inputs thoracic CT images to automatically extract 23 features of air volume variation and distribution within the lung over respiration cycle. Relationships between features and pulmonary function test (PFT) measurements were developed which indicated strong correlation. Moreover, the discriminatory power of all features were examined using sequential feature selection algorithm in both forward and backward directions. For classification, 12 features with the most discriminatory power were selected to train a Naive Bayes classifier. The study included lung inspiratory/expiratory CT images and PFT measurements of 69 subjects, including 13 normal and 56 COPD patients with various severity stages. The performance of the classifier was evaluated using leave-m-out cross-validation method with m=7. Results obtained in this investigation showed an overall accuracy of over 84% which demonstrates its effectiveness in determining COPD stage merely based on CT images and without using PFT measurements. This demonstrates the proposed expert systems potential as a clinically viable image-based COPD diagnosis method.