Classification of Emphysema Patterns in Computed Tomography Based On Gabor Filter

Emphysema is a type of chronic obstructive pulmonary disease (COPD) affecting millions of people worldwide. Patients with emphysema typically have breathing difficulty. Early detection using Computed Tomography (CT) scan image can save many of the emphysema patients life. Furthermore, it helps the medical practitioners in planning suitable treatments for patients. The CT scan of human lungs are commonly taken from 3 different directions; center, bottom and top. The images obtained from different slices are then used by radiologist to identify normal or abnormal tissues. Computer-aided diagnosis (CAD) has becomes part of routine clinical work for assisting radiologist in detection of abnormal tissue in many screening sites and hospitals. One of the main processing technique in CAD is texture classification and analysis. In this research, a Gabor-based emphysema classification algorithm is proposed. Gabor filter offer the advantage of multi-resolution and multi-orientation properties and is optimal for measuring local spatial frequencies. In essence, the Gabor transform is performed by applying Gaussian masks prior to the discrete wavelet transform. The extracted feature from the Gabor filter is in the form of local energy calculated at different scale and orientation. The proposed emphysema classification algorithm involves four aspects, image pre-processing, feature extraction, matching (classification), and decision making. In the classification stage, the k-NN classifier is used to classify the CT images to two different classes which are Normal Tissue (NT) and Abnormal Tissue; Centrilobular Emphysema (CLE) and Paraseptal Emphysema (PSE). The proposed algorithm is evaluated using k-fold cross validation technique and its performance is shown to produce low misclassification rate of 0.01%.