SU-GG-I-100: Development of Computerized Method for Automated Detection of Mesothelioma on 3D Thoracic CT

Abstract

Purpose: To develop an automated detection and segmentation scheme for the evaluation of pleural thickening on 3D thoracic CT in order to assist the early detection and follow‐up of mesothelioma. Method and Materials: Our database consists of normal 20 thoracic CTimages and abnormal 18 images including pleural plaques, calcifications and mesothelioma. In the first step, 3D region growing technique was applied to original CTimages in order to segment lung, soft tissue, and bone regions. Second, center lines of ribs were recognized by voting the curvature information. Then, calcifications were detected by excluding rib regions from segmented bone regions. In the third step, spherical shell filtering was applied to extract soft tissue regions for detection of initial pleural plaque candidates. In the fourth step, pleural plaque regions were determined by excepting mediastinum and under diaphragm regions from extracted pleural plaque candidates. Finally, cases with pleural plaques more than 10 % of total pleural volume were classified as abnormal. We made a quantitative evaluation by comparing the segmentation result with the respiratory physician's gold standard. Results: Our method was tested with a dataset of 38 CTimages. As the detection result, all 18 abnormal cases were correctly detected with 3 false positive cases (sensitivity: 100 %, specificity: 85 %). As the segmentation result, the average coincidence degree was 63.5 %. (coincidence degree = A∩B / A B, A: pleural plaque region extracted automatically; B: pleural plaque region extracted manually by respiratory physician) Conclusion: We have developed a computerized method for the automated measurement of mesothelioma on thoracic CTimages. Our method will be useful for the early detection and follow‐up of mesothelioma on thoracic CTimages.