Development of a Deep Learning-based extending the effective field of view in cone-beam CT

• Project/Area Number: 22K15804
• Research Category: Grant-in-Aid for Early-Career Scientists
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 52040:Radiological sciences-related
• Research Institution: Kyoto University
• Principal Investigator: Hideaki Hirashima 京都大学, 医学研究科, 特定助教 (10848229)
• Project Period (FY): 2022-04-01 – 2024-03-31
• Project Status: Completed (Fiscal Year 2023)
• Budget Amount: ¥2,340,000 (Direct Cost: ¥1,800,000、Indirect Cost: ¥540,000); Fiscal Year 2023: ¥910,000 (Direct Cost: ¥700,000、Indirect Cost: ¥210,000); Fiscal Year 2022: ¥1,430,000 (Direct Cost: ¥1,100,000、Indirect Cost: ¥330,000)
• Keywords: 放射線治療 / 医学物理 / AI / CBCT / 適応放射線治療

Outline of Research at the Start

治療日毎の体内臓器変化に対応し治療計画を変更する即時適応放射線治療（即時ART）は，日々変化する腫瘍や危険臓器の状態を考慮した線量分布を投与可能である．しかし，汎用的な治療装置では腫瘍や危険臓器の画像取得時に体輪郭欠損や画質劣化といった大きな課題があり，即時ARTを実施する際の輪郭描出や治療計画の障害となっている．本研究では，汎用的な治療装置を用いた即時ARTを実現するために，体輪郭欠損cone-beam CTの画質改善，及び，体輪郭補完のための画像再構成技術を開発する．

Outline of Final Research Achievements

This study examines the application of cone-beam CT (CBCT) in radiation therapy, which contains up-to-date information about the patient’s body and can be used to recalculate dose distributions for treatment planning and monitoring. However, the limited field of view (FOV) of CBCT results in a lack of body contours, which makes accurate dose calculation difficult. The aim of this study was to extend the CBCT FOV using deep learning to complete sinograms.
Three models were created using the pix2pix deep-learning algorithm: a sinogram-based model and a CT-based model, each using different image pairings.This study indicate that deep learning-based sinogram completion using CT-based modelis a viable method for extending the FOV in CBCT with missing body contours.

Academic Significance and Societal Importance of the Research Achievements

本研究では，pCTを用いた深層学習に基づくサイノグラム補完を行うことで，CBCTの有効視野（FOV）拡張を目指した．サイノグラムを補完するモデルが，画像自体を補完するモデルよりも最大値，中央値，最小値の全てにおいてMAEとRMSEの値が小さく，SSIMの値が大きかった．サイノグラムを補完するモデルが画像自体を補完するモデルよりもFOVを拡張できていた理由は，サイノグラムの情報の連続性が深層学習の学習や予測において有利に働いたからだと考えられる．本研究成果により，狭いFOVを有するCBCTを利用した正確な線量分布の計算が可能になることが期待される．

Research Products

• [Journal Article] Dosimetric verification of four dose calculation algorithms for spine stereotactic body radiotherapy (2023)
  • Author(s): Hirashima Hideaki、Nakamura Mitsuhiro、Nakamura Kiyonao、Matsuo Yukinori、Mizowaki Takashi
  • Journal Title: Journal of Radiation Research Volume: 65 Issue: 1 Pages: 109-118
  • DOI: 10.1093/jrr/rrad086
  • Peer Reviewed
• [Journal Article] Evaluation of generalization ability for deep learning‐based auto‐segmentation accuracy in limited field of view CBCT of male pelvic region (2023)
  • Author(s): Hirashima Hideaki、Nakamura Mitsuhiro、Imanishi Keiho、Nakao Megumi、Mizowaki Takashi
  • Journal Title: Journal of Applied Clinical Medical Physics Volume: e13912 Issue: 5 Pages: 1-9
  • DOI: 10.1002/acm2.13912
  • Peer Reviewed / Open Access
• [Presentation] 体輪郭欠損医用画像に対する深層学習による体輪郭復元法の開発 (2023)
  • Author(s): 有本昂平，中村光宏，平島英明，中尾恵，吉村通央，溝脇尚志
  • Organizer: 第36回高精度放射線外部照射部会学術大会
• [Presentation] Auto-Segmentation for Limited Field of View CBCT in Male Pelvic Region Using Deep Learning Method. (2022)
  • Author(s): Hirashima H, Nakamura M, Imanishi K, Nakao M, Mizowaki T.
  • Organizer: 64th AAPM annual meeting
  • Int'l Joint Research
• [Presentation] Restoration of Body Outside Field-Of-View On CT Images Using Cycle-Consistent Generative Adversarial Networks. (2022)
  • Author(s): Arimoto K, Nakamura M, Hirashima H, Nakao M, Mizowaki T.
  • Organizer: 64th AAPM annual meeting
  • Int'l Joint Research