新生児頭部MR画像における輝度値値分布と脳形状に基づく大脳領域抽出法

摘要

新生児脳疾患の診断法には，MR（magnetic resonance）画像を用いた脳容積，脳表面の計測が有効であるが・新生児を対象とした計算機による脳領域自動抽出法は少ない．本文では，新生児脳MR画像を対象とした脳領域抽出法を提案する．提案法はファジィ動的モデルを用いて脳領域を抽出し，頭蓋形状を基に抽出結果を評価することで抽出パラメータを更新する．脳領域の抽出，評価・パラメータ更新を繰り返すことで，頑健性が高く，精度が高い脳領域抽出が可能となる．初期パラメータはペイジアン識別を用いて求める．本提案法を34名の新生児被験者（修正齢-2週1日から2年5カ月）に適用した結果，sensitivityが98.1%，hlse-positive rateが27.9%と，高精度に脳領域を抽出できた．%For diagnosis of the neonatal cerebral disorders, it is useful to measure cerebral volume and surface area using magnetic resonance (MR) image. To measure cerebral volume and surface area, a brain segmentation method is required. However, there are few studies on automated neonatal brain segmentation. This study proposes an automated brain segmentation method for neonatal brain. The proposed method segments the cerebral region by using fuzzy deformable model. Then, the segmented area is evaluated with respect to the shape of whole head, and segmentation parameters of fuzzy deformable model are updated. Segmentation, evaluation and parameter update processes are iterated until satisfying predefined condition. The initial parameters are selected using Bayesian classification. Experimental results in 34 neonatal subjects (revised age between -2weeks and 1day and 2 years 5 months) showed that the brain region was segmented with sensitivity 99.4% and false-positive rate of 29.6%.