気胸変形の学習に基づく肺部分形状メッシュの可変形位置合わせ

Abstract

術中気胸は回転を含む大変形であり, 手術時に撮像可能なCone-beam CT (CBCT) は撮像領域が限られるため, 含気/虚脱肺間の正確な対応を得ることが難しい. 本研究では, 手術時に取得可能なCone-beam CT画像に含まれる肺の部分形状を対象に, 気胸変形の統計的性質の学習に基づいた可変形メッシュ位置合わせ方法を提案する. 10例の含気/虚脱肺のCBCT画像を対象に位置合わせ精度を確認する実験を行い, 従来手法と比較してより誤差が小さく安定な位置合わせが達成されることを確認したので報告する.

Intraoperative pneumothorax is accompanied by large deformation including rotation. As intraoperative cone-beam CT (CBCT) images have a limited imaging area, so it is difficult to obtain an accurate correspondence between inflated and deflated lungs. In this study, we propose a deformable mesh registration method based on learning the statistical characteristics of pneumothorax deformation, targeting the partial shape of the lung included in paired CBCT images in the inflated and deflated states. We have evaluated registration accuracy for 10 CBCT images of inflated/deflated lungs, and confirmed that the proposed method achieves stable registration with smaller errors compared to existing methods.