Automated detection of lung nodules in CT scans: Preliminary results

摘要: We have developed a fully automated computerized method for the detection of lung nodules in helical computed tomography(CT) scans thorax. This is based on two-dimensional and three-dimensional analyses image data acquired during diagnostic CT scans.Lung segmentation proceeds section-by-section basis to construct segmented volume within which further analysis performed. Multiple gray-level thresholds are applied create series thresholded volumes. An 18-point connectivity scheme used identify contiguous structures each volume, those that satisfy criterion selected as initial nodule candidates. Morphological features candidate. After rule-based approach greatly reduce number candidates corresponds non-nodules, remaining merged through linear discriminant analysis. The was database 43 thoracic scans. Receiver operating characteristic (ROC) evaluate ability classifier differentiate correspond actual from false-positive area under ROC curve this categorization task attained value 0.90 leave-one-out-by-case evaluation. yielded an overall sensitivity 70% with average 1.5 detections per section when complete 43-case database. A corresponding 89% 1.3 achieved subset 20 cases contained only one or two case.