竹島 昇吾¹, 桝田 晃司¹, BOSSARD Antoine¹, 渡辺 弘之²

Shogo TAKESHIMA¹, Kohji MASUDA¹, Antoine BOSSARD¹, Hiroyuki WATANABE²

¹東京農工大学大学院生物システム応用科学府, ²東京ベイ・浦安市川医療センター

¹Graduate School of Bio-Applications and Systems Engineering, Tokyo University of Agriculture and Technology, ²Tokyo Bay Urayasu Ichikawa Medical Center

キーワード : left ventricle, motion vector, automatic recognition, ejection fraction, four chamber view

超音波断層像における心機能の計測手法では，現状では内腔形状を手動によって数点クリックするなどの初期設定が必要である．この操作を省略可能な方法として，我々は運動ベクトルの交点検出アルゴリズムによる短軸断面における左室腔形状の自動認識手法を開発してきたが，今回は新たな処理を追加し，同様の認識を四腔断面にも適用可能とした．まず，心臓が描出された四腔断面像に対して複数方向からの輝度走査により左室の概形を取得し，左室以外の処理対象から除外した．そして運動ベクトルの交点領域から導かれる関心領域を展開し，断層像の輝度分布から僧帽弁輪部を特定し，得られた左室腔の特徴点にB-Spline補間を適用し，左室腔形状を決定した．この手法では，断層像中に左室が含まれていれば操作者側での初期設定が一切不要である．実際の臨床で得られた50例の疾患心の超音波断層像に対して適用し評価を行った結果，画質が悪い14例を除いて89％の認識率を得ることができた．さらに，抽出された左室腔形状の時間変化からEjection Fractionを自動的に計測したところ，従来の計測方法との相関係数は0.82となった．これらの結果は，本システムが再現性や時間効率を向上させるために臨床上有用であると推測された．

Since examinations with echography largely depend on the expertise and experience of the operator, many kinds of techniques for recognizing the left ventricular (LV) cavity have been developed. However, most of these techniques require initial settings to indicate the initial region or points of interest of the examiner. Thus, an automatic diagnosis support system without initial settings has the potential not only to save time when analyzing large amounts of data but also to reduce the burden of doctors or sonographers. Therefore, we have developed automatic recognition software for the LV cavity that does not require initial settings. The software consists of two processes. The first process is the automatic detection of the inner area of the LV cavity, which includes calculation of the motion vectors from the ventricular wall and calculation of intersection points from multiple combinations of vectors. The second process is the automatic recognition of the LV cavity, which includes the approximation of the shape of the LV to ellipsoid by centering the gravity point of the intersection points, and the determination of the LV cavity without the mitral valve and the nipple muscle. We have applied this algorithm to 50 diseased hearts. The proposed method correctly recognized LV cavities with an 89% recognition rate in 36 cases, where 14 severe cases were excluded. Furthermore, the results of ejection fraction calculation agreed with those yielded by the conventional method performed by professional sonographers, with a correlation coefficient of 0.82.