Radiotherapy and oncology : journal of the European Society for Therapeutic Radiology and Oncology · 2025
Automated vertebral identification and localization for enhanced radiotherapy patient setup.
СтатьяНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Zhang Jie, Pan Hailun, Wang Fakai, Jia Lecheng, Chen Jiayi, Yan Fuhua, Zhang Yibin, Yang Yingli
Аннотация
BACKGROUND AND PURPOSE: Vertebral bodies are critical landmarks for image-based patient positioning during radiotherapy (RT). However, manual identification of vertebral bodies can be laborious and a source of error, potentially leading to treatment mistakes. This study demonstrated an automated technique for vertebral identification and localization in images with varying quality and field of view (FOV), aiming to streamline the positioning process and minimize the risk of patient misalignment.
MATERIALS AND METHODS: This retrospective study employed an nnU-Net-based model for automated vertebral identification and localization. Training was performed on 1,053 datasets (993 public datasets: SpineWeb, Verse19, Verse20, Spine1K; 60 clinical on-board CT scans from Infinity® and TomoTherapy®). Testing included 688 public datasets, 155 clinical on-board CT scans (Ethos®, Infinity®, TomoTherapy®, TrueBeam®, Trilogy®), and 50 clinical planning CTs (Brilliance CT Big-Bore-Oncology®). A strategic four-step post-processing procedure was developed to enhance accuracy, considering the anatomical characteristics of vertebral structures and vertebral abnormalities. Evaluation metrics included identification rates, mean localization errors, and their standard deviations.
RESULTS: The method achieved high identification rates of 97.99 % with a mean localization error of 1.64 ± 1.23 mm on public datasets and 99.76 % with a mean error of 1.74 ± 1.36 mm on clinical datasets. Refining traditional 20 mm accuracy thresholds, new section-specific thresholds were established 10 mm for cervical, 15 mm for thoracic, and 19 mm for lumbar vertebrae.
CONCLUSIONS: This automated approach offers an accurate and widely applicable model for vertebral identification and localization. It has the potential to enhance RT setup workflows and serve as a valuable clinical tool.
MATERIALS AND METHODS: This retrospective study employed an nnU-Net-based model for automated vertebral identification and localization. Training was performed on 1,053 datasets (993 public datasets: SpineWeb, Verse19, Verse20, Spine1K; 60 clinical on-board CT scans from Infinity® and TomoTherapy®). Testing included 688 public datasets, 155 clinical on-board CT scans (Ethos®, Infinity®, TomoTherapy®, TrueBeam®, Trilogy®), and 50 clinical planning CTs (Brilliance CT Big-Bore-Oncology®). A strategic four-step post-processing procedure was developed to enhance accuracy, considering the anatomical characteristics of vertebral structures and vertebral abnormalities. Evaluation metrics included identification rates, mean localization errors, and their standard deviations.
RESULTS: The method achieved high identification rates of 97.99 % with a mean localization error of 1.64 ± 1.23 mm on public datasets and 99.76 % with a mean error of 1.74 ± 1.36 mm on clinical datasets. Refining traditional 20 mm accuracy thresholds, new section-specific thresholds were established 10 mm for cervical, 15 mm for thoracic, and 19 mm for lumbar vertebrae.
CONCLUSIONS: This automated approach offers an accurate and widely applicable model for vertebral identification and localization. It has the potential to enhance RT setup workflows and serve as a valuable clinical tool.