サポートベクターマシンを用いた肺腫瘍の鑑別

Classification of pulmonary nodules using support vector machine

Creators Yamada Nami

Creators Tachibana Rie

Creators Kido Syoji

Accurate segmentation of small pulmonary nodules (SPNs) is an important technique for volumetric doubling time estimation and for feature characterization of the diagnosis of SPNs. Therefore, we have developed an automated volumetric segmentation algorithm of SPNs on thoracic CT images. This paper presents quantitative evaluation of the shape of the segmented SPNs on thoracic CT images, and also presents quantitative estimation of the growth rate of SPNs by use of repeated thoracic CT images. To quantitatively evaluate the performance of our algorithm, we employed the coincident rate. This coincident rate was calculated with both the computerized segmented region of an SPN and the region traced by a chest radiologist on axial, coronal MPR, sagittal MPR images. As the result of 90 selected cases, the mean value of the total coincident rate was 0.75 +/- 0.081. We also estimated the performance of our algorithm by use of the doubling time of SPNs on repeated CT images. The doubling time of SPNs obtained from computerized segmented regions closely correlated with the image findings reported by radiologists. From these results, we consider that our algorithm is useful for quantitative estimation for the shape and the growth rate of SPNs on repeated CT in daily clinical practice.

[ISSN]0387-9232

[NCID]AN00031668