三次元造影CT像上の膵臓領域抽出のための二段階セグメンテーション

摘要

We have developed the segmentation method for multiple organs including pancreas from three dimensional abdominal CT images. Although the segmentation results of large organs such as liver and spleen were acceptable, the results of small organs such as pancreas were far from enough for CAD system. In this paper, we propose the two stage process to extract pancreas where liver and spleen are extracted in advance and the extracted regions are used to normalize the position of pancreas. In the pre process, the input three phase CT images (early, portal and late phase images) are aligned. Then the proposed method carried out liver and spleen segmentation as the first stage followed by normalization of pancreas position using the extracted regions. Second pancreas is roughly extracted based on the maximum a posterior (MAP) method using the probabilistic atlas and it performs rule-based analysis of the connected components and morphological operations with voxel-wise classification function based on Mahalanobis distance. This paper shows the results of applying the proposed method to three phases CT images taken from 36 cases and the discussion about the effective based on the results.%著者らはこれまで三次元腹部CT像を対象に，膵臓領域を含む複数臓器領域の同時抽出手法を開発してきた．そこでは，肝臓や牌臓領域などで高精度な抽出に成功したが，体積が小さく，形状が複雑な膵臓領域の抽出精度には問題が残されていた．本研究では，膵臓に隣接する肝臓と肺臓領域を予め抽出し，それらを用いて標準化を行った後で，膵臓領域を抽出する段階的な領域抽出手法を提案する．具体的には，まず．放射基底関数を用いた3時相の位置合わせ，肝臓，牌膿の抽出とそれを用いた標準化を行う．その後，確率アトラスを用いた最大事後確率法に基づく臓器領域の粗抽出を行い，ルールベースによる領域判別と，マハラノビス距離に基づく画素単位の判別機能を持つモルフォロジー処理により膵臓領域の精密な抽出を行う．実験では，提案する手法を36症例の3時相の造影CT像に適用し，その有効性について考察した．