4時相の3次元腹部CT像を用いた膵臓領域抽出手法の開発

摘要

本稿では，異なる4時相において撮影されたCT健から膵臓領域を抽出する手法を提案する．初めに，各CT像に対してメディアンフィルタによる平滑化処理を施した後，非剛体レジストレーションにより4時相のCT像の位置を合わせる．次に早期相像と晩期相橡から2次元濃度億ヒストグラムを，また，非造影像から1次元濃度値ヒストグラムを作成し，これらのヒストグラムから膵臓領域に対応するCT債分布を推定する．そしてマハラノビス距離を用いて各画素にラベルを割り当て，膵臓候補領域を粗抽出する．さらに，膵臓領域は，候補領域内の膵臓以外の領域との境界領域に比べてCT値の分散が低いことを利用し，分散情報を用いた領域拡張により膵臓に対応する領域を抽出する．提案手法を4時相腹部3次元CT像21例に対して適用した結果．10例において良好な抽出結果が得られたが，8例においては，抽出された膵臓領域の一部に過抽出および抽出不足が存在した．%This paper presents a method for extracting pancreas regions from four-phase contrasted CT images taken at different contrast timing. First of all, a median filtering and a nonrigid registration are applied to each image for reducing speckle noise and motion artifact. Then, a 2D histogram is created from the early- and late-phase CT images. We first estimate the CT value distribution corresponding to the spleen region from the histogram. Organ labels are assigned to each voxel by using Mahalanobis distance measure. We extract voxels having the label of the spleen and then estimate the pancreas region from the 2D histogram of the early- and late-phase images and ID histogram of non-contrasted image, where the spleen region has been eliminated. A candidate region of the pancreas is obtained by the same way applied to the spleen. In addition, since the pancreas region has lower variances of CT values than those of boundary regions in the candidate region, we utilize a region growing method which uses variances of CT values in a growing condition, to extract low variance region as the pancreas. Applying this method to 21 cases of four-phase abdominal CT images, pancreas regions for 10 cases were extracted satisfactorily, however, the regions for 8 cases were extracted including FP and FN regions.