Segment Segmentation in Lung CT Images - Preliminary Results

Segmentation of lung anatomical parts like fissures and bronchopulmonary segments are of clinical importance in interpreting and assessing the pathologies present in the lungs. Segmentation of fissures still remains a challenging task as the selection of the segmentation algorithm decides the accuracy of the fissure segmentation. We conducted experiments on the existing segmentation methods and propose an automatic delineation method for fissure segmentation in High Resolution Computed Tomography images of the lungs. The method applies Gray Level Co-occurrence Matrix measures to enhance and detect the fissures. The feature set calculated from GLCM values are applied to a supervised enhancement filter. A KNN classifier is then trained to classify the pixels belonging to fissures by correlating the human observed and automated feature set calculated. A marker controlled watershed algorithm is applied to segment the bronchopulmonary segments as the next step using detected fissures. The method is experimented with 5 test images and the accuracy obtained is 57% comparable to the radiologist observation.