大腸内視鏡治療誘導のためのRecurrent Neural Networkを用いた大腸内視鏡トラッキング手法の開発

抄録

This paper presents a colonoscope tracking method for colonoscope navigation. Understanding colonoscope position in the colon during colonoscopic examinations is difficult due to unclear camera view. Computerized colonoscope navigation based on an accurate colonoscope position identification method in the colon is required. The position identification is called as tracking. We propose a colonoscope tracking method using a colon deformation estimation technique. Colonoscope tracking is difficult because the colon largely deforms during colonoscope insertions. We estimate colon deformation using recurrent neural network-based estimation method. The colonoscope tip position is mapped to a CT image coordinate system to combine the position with a colon shape and polyp positions. We evaluated the colonoscope tracking accuracy in a colon phantom study. From our experiments, tracking errors were small compared to a previous method. The proposed method has potential to perform tracking in clinical fields.