講演抄録/キーワード |
講演名 |
2022-11-18 14:00
部分X線画像の外挿による全身筋骨格系構造の予測 ○チョウ ウェイキ・谷 懿・大竹義人・崇風まあぜん(奈良先端大)・上村圭亮(阪大)・高尾正樹(愛媛大)・明石敏昭(順天堂大)・森 健策(名大/NII)・合田憲人(NII)・菅野伸彦(阪大)・佐藤嘉伸(奈良先端大) MICT2022-38 MI2022-67 |
抄録 |
(和) |
Image defects and partial disorders are common problems in medical imaging. Image inpainting and extrapolation are helpful techniques to restore missing image information for many clinical applications, including analyzing organs or tissues that are not scanned during imaging processing. Many image restoration algorithms have been proposed for general images. However, numerous challenges still exist in restoring medical images, such as domain shifting and limited datasets. Conventional methods for medical image restoration only focused on recovering small regions within a given image, ignoring clinical demands for extrapolation. In this study, we proposed a method based on the transformer for restoring an X-ray image with large regions, which supports whole-body restoration from a partial region, using mutual conversion between an X-ray image and a digitally reconstructed radiograph. Our method combined an extrapolation network and a style transfer network, simultaneously achieving the inpainting and extrapolating of an X-ray image under a limited dataset. To the best of our knowledge, we are the first to achieve whole-body restoration from an X-ray image. We conducted 1) quantitative and qualitative experiments on the extrapolation and style transfer models and 2) bone mineral density (BMD) estimation experiments from extrapolated X-ray images generated by the proposed method. We used the predicted and ground-truth BMD correlation to evaluate our model's effectiveness, which achieved PCC of 0.374 and 0.534 in DXA-measured and QCT-measured BMD, respectively, demonstrating the high clinical potential of analyzing missing regions using the proposed method. |
(英) |
Image defects and partial disorders are common problems in medical imaging. Image inpainting and extrapolation are helpful techniques to restore missing image information for many clinical applications, including analyzing organs or tissues that are not scanned during imaging processing. Many image restoration algorithms have been proposed for general images. However, numerous challenges still exist in restoring medical images, such as domain shifting and limited datasets. Conventional methods for medical image restoration only focused on recovering small regions within a given image, ignoring clinical demands for extrapolation. In this study, we proposed a method based on the transformer for restoring an X-ray image with large regions, which supports whole-body restoration from a partial region, using mutual conversion between an X-ray image and a digitally reconstructed radiograph. Our method combined an extrapolation network and a style transfer network, simultaneously achieving the inpainting and extrapolating of an X-ray image under a limited dataset. To the best of our knowledge, we are the first to achieve whole-body restoration from an X-ray image. We conducted 1) quantitative and qualitative experiments on the extrapolation and style transfer models and 2) bone mineral density (BMD) estimation experiments from extrapolated X-ray images generated by the proposed method. We used the predicted and ground-truth BMD correlation to evaluate our model's effectiveness, which achieved PCC of 0.374 and 0.534 in DXA-measured and QCT-measured BMD, respectively, demonstrating the high clinical potential of analyzing missing regions using the proposed method. |
キーワード |
(和) |
画像の外挿 / X線画像 / トランスフォーマー / 骨密度(BMD)推定値 / / / / |
(英) |
Image Extrapolation / X-ray Image / Transformer / Bone Mineral Density (BMD) Estimation / / / / |
文献情報 |
信学技報, vol. 122, no. 265, MI2022-67, pp. 24-28, 2022年11月. |
資料番号 |
MI2022-67 |
発行日 |
2022-11-11 (MICT, MI) |
ISSN |
Online edition: ISSN 2432-6380 |
著作権に ついて |
技術研究報告に掲載された論文の著作権は電子情報通信学会に帰属します.(許諾番号:10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
PDFダウンロード |
MICT2022-38 MI2022-67 |
|