膝のMR画像における関節軟骨輪郭の半自動抽出法

摘要

Knee motion analysis has many clinical applications such as diagnosis of knee diseases and evaluation of postoperative knee functions. For analyzing knee motion in three-dimensions, a 3-D model based approaches were introduced recently by combining a fluoroscopic image-sequence of knee motion and a 3-D model of related knee bones. In order to expand such analysis into that including the articular cartilage, we need a 3-D model of the cartilage as well as bones. Then this paper proposed a semi-automatic extraction of cartilage contour in MR images. This extraction method has an advantage that a contour of the cartilage partly unclear in MR image can be estimated iteratively by using a function of evaluating the overall difference between present contour and that estimated from its neighboring gray distribution, from an initial contour roughly digitized manually. To evaluate its validity, we carried out several experiments using two different MR images of the knee. All empirical results suggested that our method would have a potential of easily extracting an accurate contour of the cartilage with high reproducibility.%膝運動の解析には，膝疾患の病態解明や術後の快復状態の把握など，多くの臨床応用がある.3次元で膝運動を解析するために，最近，透視X線動画像と3次元骨モデルを重ね合わるアプローチが用いられている.この手法を「関節軟骨」を含めた動態解析に拡張するためには，骨のみならず軟骨の3次元モデルが必妻となる.そこで本論文では，「MR画像における関節軟骨輪郭を半自動抽出する方法」を提案する.この抽出法には，部分的に不鮮明な関節軟骨輪郭を，手入力した初期輪郭から反復的に抽出することができる利点がある.反復の評価関数は，輪郭付近の濃度値を用いて更新前後の輪郭の違いを評価する.半自動抽出された輪郭の妥当性を評価するために，我々は2枚の膝のMR画像を用いていくつか実験を行った.すべての実験結果は，提案した輪郭抽出法が膝の関節軟骨の正確な輪郭を高い再現性で，かつ簡単に抽出できる可能性があることを示唆していた.