Automatic detection of large pulmonary solid nodules in thoracic CT images

摘要: Purpose: Current computer-aided detection (CAD) systems for pulmonary nodules in computed tomography (CT) scans have a good performance relatively small nodules, but often fail to detect the much rarer larger which are more likely be cancerous. We present novel CAD system specifically designed solid than 10 mm. Methods: The proposed pipeline is initiated by three-dimensional lung segmentation algorithm optimized include large attached pleural wall via morphological processing. An additional preprocessing used mask out structures outside space ensure that and parenchymal similar appearance. Next, nodule candidates obtained multistage process of thresholding operations, both smaller candidates. After segmenting each candidate, set 24 features based on intensity, shape, blobness, spatial context computed. A radial basis support vector machine (SVM) classifier was classify candidates, evaluated using ten-fold cross-validation full publicly available image database consortium database. Results: reaches sensitivity 98.3% (234/238) 94.1% (224/238) at an average 4.0 1.0 false positives/scan, respectively. Conclusions: authors conclude dedicated can identify vast majority highly suspicious lesions thoracic CT with number positives.