GPU-Based Airway Tree Segmentation and Centerline Extraction

摘要: Lung cancer is one of the deadliest and most common types in Norway. Early precise diagnosis crucial for improving survival rate. Diagnosis often done by extracting a tissue sample lung through mouth throat. It difficult to navigate because complexity airways inside reduced visibility. Our goal make program that can automatically extract map Airways directly from X-ray Computer Tomography(CT) images patient. This complex task requires time consuming processing. In this thesis we explore different methods CT images. We also investigate parallel processing usage modern graphic units speeding up computations. rate several terms reported performance possibility The best rated method implemented framework called Open Computing Language. results shows our implementation able large parts Airway Tree, but struggles with smaller deviate perfect circular cross-section. process full scan using less than minute units. very general other tubular structures as well. To show run on Magnetic Resonance Angio dataset finding blood vessels brain achieve good results. see lot potential structures. noise handling tubes cross-sectional shape. believe be improved another ridge traversal centerline extraction step. Because local greedy algorithm, it terminates prematurely due image artifacts.