歯科金属アーチファクト低減向け3次元敵対的生成ネットワークの性能評価

Abstract

本研究では歯科金属アーチファクト低減を実現する3次元敵対的生成ネットワークを構築した．915の実患者3次元CT画像を歯科金属の有無に基づいて分類した金属アーチファクトデータベースを構築し，設計した正則化付き損失関数に基づいて敵対的トレーニングを行った．複数の歯科金属から生成される広範囲な画素欠損を伴うアーチファクトを対象に，アーチファクト低減及びCT値復元性能の評価を試みたので報告する． In this study, we propose a method for metal artifact reduction in real CT images based on unsupervised image transfer. We do not suppose the existence of training data with reduced artifacts or synthesized images of metal artifacts, and focus on a problem to obtain domain transfer between clinical CT images with and without dental metals. Based on the concept of CycleGAN, a novel loss function that reduces metal artifacts while preserving their CT values was designed. A adversarial training framework considering three-dimensional (3D) anatomical structures and 3D distribution of metal artifact was developed. This presentation reports the preliminary results of metal artifact reduction effectively performed by domain transfer learnt from 96 3D-CT images.