Automated vessel tree segmentation in lungs

Automated vessel tree segmentation in the presence of interstitial lung disease in Multidetector Computed Tomography (MDCT) is an important task in the development of Computer Aided Diagnosis (CAD) scheme. The accuracy of automated vessel tree segmentation aims to improve the accuracy in CAD scheme. The development of CAD scheme is for quantification, detection and diagnosis of ILD. Vessel tree segmentation deal with the normal lung parenchyma and ILD affected lung parenchyma. In case of ILD affected lung, initial stage of the method identifies the vessel tree volume candidate by utilizing the combination of three-dimensional multiscale enhancement filtering and Expectation Maximization segmentation algorithm. The second stage of the method is texture based analysis for accurate segmentation. The proposed method utilizes Otsu threshold to segment the lung field and Fuzzy Support Vector Machine classifier is also enhanced to improve the traditional SVM by adding fuzzy membership to training sample to indicate degree of membership of this sample to different class and accuracy of SVM. Consequently it reduces noises and outliers in data and enhances performance. Performance of vessel tree accuracy is evaluated by means of Area Overlap (AO), True Positive Fraction (TPF) and False Positive Fraction (FPF) metrics. Automated vessel tree segmentation is expected to improve performance as compared to other techniques.