皮下脂肪領域の除去によるCT画像の仮想展開と腹直筋領域の自動認識

摘要

We had proposed an automated method for recognizing skeletal muscle in torso X-ray CT images using skeletal positional information and shape model of its muscle, and confirmed that the method was effective in recognizing the muscles. The previous method recognized the muscle's connecting points on the virtually unfolded image of the human body as landmarks and fitted a shape model based on the positional information obtained from the landmarks, to recognize the muscle region automatically. However, there was a variation in recognition accuracy at the site where the surface shape is greatly different depending on the difference in physique such as an abdominal region. In this paper, we have improved the virtual image-unfolding technique. The proposed method removed the subcutaneous fat area beforehand because of its significant difference by physique between patients, and applied the remaining regions for recognition of the muscle. We applied this methodology to 10 patients with no evidence of abnormalities to the rectus abdominis muscle. We achieved the 89% concordance with manual segmentation and confirmed the effectiveness of the muscle polyventer shape modeling and recognition.%我々はこれまでに，非造影体幹部X線CT画像から骨格の位置情報と骨格筋の形状モデルを用いた骨格筋の自動認識手法を提案し，解剖学に基づく特徴点の検出と骨格筋の形状モデルは骨格筋の自動認識に有効な手法であると確認した.従来法では，人体を仮想的に切り開いた展開画像上において，骨格から取得した骨格筋の付着箇所をランドマークとして認識し，ランドマークから得られる位置情報を元に形状モデノレを当てはめ，骨格筋領域の自動認識を行った.しかしながら，腹部のような体格の差異による表面形状が大きく異なる部位では，認識精度にばらつきが生じた.本研究では，この展開画像生成法を改良した.ここでは，患者間で差異が大きく，体格に大きな影響を与える皮下脂肪領域を事前に除去し，生成した展開画像を腹直筋の認識に適用した.腹直筋に所見のない10症例に本手法を適用し，平均一致率89%の結果が得られ，本手法は多腹筋形状のモデル化および，腹直筋の認識に有効であることが確認できた.