Lung affected Patterns of Automatic Vessel segmentation in MDCT using Decision Tree Classification

Pulmonary vascular tree segmentation is gaining importance since it is one of the fundamental basis for different applications, such as the detection of interstitial pneumonia (IP), pulmonary emboli etc. Such an application will require an accurate and reliable segmentation of pulmonary vessels. The accuracy of this preprocessing stage is bound to influence the accuracy in computer aided diagnosis (CAD) of IP patterns. While lot many algorithms aimed at improving accuracy of lung segmentation, vessel tree segmentation is still an open research issue. In this paper an automated vessel tree segmentation algorithm with high accuracy is proposed in presence of pathologies affecting lung parenchyma. The initial stage accounts for a vessel enhancement filtering, which uses second order local structure of an image (Hessian) with vessel ness measure obtained on the basis of all eigen values of the Hessian. Followingly texture based refinement using 3D co-occurrence matrix, which ties all textures together (a single value per feature will take into account all texture features within a lung, finally the classification is performed to correct possible over segmentation. The proposed method utilizes Decision Tree classifier which improves traditional SVM by adding membership to training sample to indicate degree of membership of this sample to different class. Consequently it reduces noises and outliers in data and enhances performance and accuracy of SVM. The performance of the proposed scheme, and of the previously reported technique, in vessel tree segmentation was evaluated by means of area overlap; true positive fraction and false positive fraction of image dataset obtained from IP affected patient scans. The method is expected to improve the performance as compared to other reported techniques.