Nodule volume change estimation in thoracic CT using sphere fitting , morphological segmentation and image registration

Three methods are presented to assess the relative volume change of a pulmonary nodule between two chest CT scans when the nodule location in both scans is provided. The first method fits a sphere around both nodules and computes the volume of dense tissue in that sphere. The second method segments both nodules using thresholding, component labeling and morphological processing. The third method applies non-rigid registration to transform the first to the second scan and applies that transformation to a segmentation of the nodule in the first scan to obtain a segmentation of the nodule in the second scan. All methods are applied to 50 nodule pairs from the VOLCANO’09 challenge. These cases are divided by a radiologist in stable and growing pairs. All methods produce lower mean volume change for the stable cases compared to the growing nodules, but the distributions overlap considerably. Moreover, the correlation between the volume change estimates produced by the three methods is modest. This shows that nodule volume change assessment is a complicated problem.