Chronic Obstructive Pulmonary Disease (COPD) is characterized by airflow limitations which involves chronic bronchitis and emphysema. Quantification of the emphysema severity is done by physicians or radiologist and typically consists of visual scoring of chest tomographic (CT) images which is subject to inter-observer variability. The purpose of this project was to study an objective automatic method using texture analysis for the detection and quantification of emphysema in patients with COPD.
The proposed method segmented the lungs using region growing and texture features were extracted after applying co-occurrence matrix algorithm to the segmented images. These texture features were used to train and test the support vector machine (SVM) classifier to distinguish between normal lung tissue and emphysematous lung tissue. This classifier was trained and evaluated using the leave-one-out algorithm with a data set of 9 non healthy patients and 2 healthy patients.
In order to validate the results of the proposed method, a comparison between them and the visual scoring of two radiologist and one physician were computed. The statistical method used for measuring the level of agreement was the quadratic weighted kappa which gave as results a fair agreement between experts and the proposed method.
The proposed method described here can quantify emphysema severity avoiding the problem of the inter-observer variability. Therefore, this method can be taken into consideration as an approach to automatic quantify emphysema lesions in patients with COPD.