Automatic Lung Nodule Registration in Serial CT Scans

摘要

To identify corresponding pulmonary nodules in serial CT scans for interval change analysis, we propose a multi-stage registration technique. Our method is composed of two main steps. First, Gross translational mismatch is corrected by the coronal and sagittal MIPs-based rigid registration. These MIP images contain a rib structure which has the highest intensity region of the chest. Second, a nodule template in initial CT scans and a search volume in follow-up CT scans are defined respectively. The template matching is performed by searching for the maximum normalized cross-correlation between the nodule template and the candidate search volumes. For fast search, gray thresholding and multi-resolution technique is applied to the search volume. As the experimental results, the hitting rate of sixty nine pulmonary nodules in fifty serial chest CT scans was 98.5% and the total processing time per subject was 3.3 seconds on average.