Monte Carlo-consistent dose prediction for clinical CyberKnife radiotherapy using a physics-and spatially-informed diffusion modelOA
Introduction:Although Monte Carlo(MC)dose calculation is the gold standard for Cyber Knife radiotherapy,its clinical integration is hindered by prohibitive computational latency arising from stochastic particle transport in complex noncoplanar and small-field geometries.To address this,we propose a physics-and spatially informed diffusion model incorporating a vision transformer(PSIDMVi T)to achieve efficient prediction of MC-consistent dose distributions.Materials and methods:A total of 251 cancer patients(117 head and neck,76 lung,and 58 liver)were retrospectively enrolled as study participants and randomly divided at the patient level into training,validation,and test cohorts at an 8:1:1 ratio.For each patient,the dataset included computed tomography(CT)images,structure sets,and clinical dose distributions calculated using both finite-size pencil beam(FSPB)and MC algorithms,with the latter serving as the ground truth.We developed a novel PSIDMViT for dose prediction that uses FSPB dose distributions as physical priors and multi-channel signed distance maps as spatial priors,with planning CT images providing complementary anatomical context.We evaluated model performance using metrics that included peak signal-to-noise ratio(PSNR),structural similarity index measure(SSIM),mean absolute error(MAE),and 3D Gamma passing rates.We also measured the prediction time to assess computational efficiency.Results:The PSIDMViT achieved significantly higher predictive accuracy than the FSPB and baseline models(P<0.05).Across all anatomical sites,the model substantially enhanced dose fidelity,with average improvements of 90%and 130%in PSNR and SSIM,respectively,and a reduction of approximately 87%in MAE.The 3D Gamma passing rates(1%/1 mm/10%)improved to 98.000%±2.500%,94.000%±1.500%,and 95.000%±1.800%for head and neck,lung,and liver cancer cases,respectively.Furthermore,the model achieved a 3.5-fold speedup over the GPU-accelerated MC baseline,reducing the total computation time from approximately 1 h to 18 min of inference time.Conclusions:The proposed PSIDMViT model effectively synergized physical,anatomical,and spatial priors,thereby providing a promising and efficient framework for high-precision dose calculation in Cyber Knife radiotherapy.
Li Mingzhu;Hui Xu;Xiao Zhang;Hongyu Lin;Shihuan Qin;He Huang;Zunhao Zhang;Yiming Ren;Mengxiao Peng;Jiapeng Li;Ruiyan Du;Wei Liu;Ying Li;Lian Zhang
Medical AI Lab,The First Hospital of Hebei Medical University,Shijiazhuang Hebei 050000,China Hebei Provincial Engineering Research Center for AI‑Based Cancer Treatment Decision‑Making,The First Hospital of Hebei Medical University,Shijiazhuang Hebei 050000,ChinaDepartment of Oncology,The First Hospital of Hebei Medical University,Shijiazhuang Hebei 050000,ChinaMedical AI Lab,The First Hospital of Hebei Medical University,Shijiazhuang Hebei 050000,China Hebei Provincial Engineering Research Center for AI‑Based Cancer Treatment Decision‑Making,The First Hospital of Hebei Medical University,Shijiazhuang Hebei 050000,ChinaDepartment of Oncology,The First Hospital of Hebei Medical University,Shijiazhuang Hebei 050000,ChinaMedical AI Lab,The First Hospital of Hebei Medical University,Shijiazhuang Hebei 050000,China Hebei Provincial Engineering Research Center for AI‑Based Cancer Treatment Decision‑Making,The First Hospital of Hebei Medical University,Shijiazhuang Hebei 050000,ChinaDepartment of Oncology,The First Hospital of Hebei Medical University,Shijiazhuang Hebei 050000,ChinaDepartment of Oncology,The First Hospital of Hebei Medical University,Shijiazhuang Hebei 050000,ChinaMedical AI Lab,The First Hospital of Hebei Medical University,Shijiazhuang Hebei 050000,China Hebei Provincial Engineering Research Center for AI‑Based Cancer Treatment Decision‑Making,The First Hospital of Hebei Medical University,Shijiazhuang Hebei 050000,ChinaMedical AI Lab,The First Hospital of Hebei Medical University,Shijiazhuang Hebei 050000,China Hebei Provincial Engineering Research Center for AI‑Based Cancer Treatment Decision‑Making,The First Hospital of Hebei Medical University,Shijiazhuang Hebei 050000,ChinaMedical AI Lab,The First Hospital of Hebei Medical University,Shijiazhuang Hebei 050000,China Hebei Provincial Engineering Research Center for AI‑Based Cancer Treatment Decision‑Making,The First Hospital of Hebei Medical University,Shijiazhuang Hebei 050000,ChinaMedical AI Lab,The First Hospital of Hebei Medical University,Shijiazhuang Hebei 050000,China Hebei Provincial Engineering Research Center for AI‑Based Cancer Treatment Decision‑Making,The First Hospital of Hebei Medical University,Shijiazhuang Hebei 050000,ChinaDepartment of Radiation Oncology,Mayo Clinic,Phoenix AZ 85054,United StatesDepartment of Oncology,The First Hospital of Hebei Medical University,Shijiazhuang Hebei 050000,ChinaMedical AI Lab,The First Hospital of Hebei Medical University,Shijiazhuang Hebei 050000,China Hebei Provincial Engineering Research Center for AI‑Based Cancer Treatment Decision‑Making,The First Hospital of Hebei Medical University,Shijiazhuang Hebei 050000,China Department of Oncology,The First Hospital of Hebei Medical University,Shijiazhuang Hebei 050000,China
医药卫生
CyberKnifeDiffusion modelMonte CarloDose calculationFinite‑size pencil beam
《Intelligent Oncology》 2026 (3)
P.4-14,11
supported by the Hebei Provincial Health Commission Medical Research Program(Grant No.:20250011,20260098 and 20260115)the Innovation&Development Medical Cooperation Program of Hengrui-Hebei(Grant No.:HR202502085)the Hebei Provincial Yanzhao Golden Terrace Talent Recruitment ProgramTop-Tier Talent Program(Grant No.:HY2025050008)the Chronic Disease Management Research Program of the Center for Capacity Building and Continuing EducationNational Health Commission(Grant No.:GWJJMB202510022202)the National Foreign Experts Program(Grant No.:S20250235)the Spark Scientific Research Program of the First Affiliated Hospital of Hebei Medical University(Grant No.:XH202515)。
评论