Medical physics · 2025
A novel dose calculation system implemented in image domain.
СтатьяНазвание и аннотация: Перевод в очереди
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.
Fan Jiawei, Zhu Xiaoling, Wang Jiazhou, Men Kuo, Dai Jianrong, Hu Weigang, Liu Zhiqiang
Аннотация
PURPOSE: Modern intensity-modulated radiotherapy, aiming to deliver an accurate dose to the planning target volume while protecting the surrounding organs at risk, is regarded as the indispensable treatment for cancer in the clinic. An accurate and efficient dose calculation algorithm is essential to support and accelerate the treatment plan optimization process.
METHOD: To improve the effectiveness of the dose calculation, deep learning methods are widely adopted to compute dose distributions from several variables mimicking the physical features required for dose calculation. However, these methods employed similar concepts and neural network inputs, and had only limited improvements to traditional dose calculation methods. In this work, we propose a new computing process, named DeepBEVdose, which is essentially distinct to the previous deep learning-based dose calculation methods. We present a novel image-domain dose calculation algorithm to automatically compute dose distributions from the computer tomography images and radiation field fluence maps. Specifically, a novel beam's eye view calculation scheme is introduced to substitute the traditional trivial ray-tracing procedure that cannot be removed in the previously published dose calculation system. Under this new calculation scheme, a generic two-dimensional convolutional neural network with minimal input requirements is sufficient to perform accurate dose calculation.
RESULTS: We demonstrate the feasibility of this approach with datasets acquired from multi-institutions with two different tumor sites (nasopharynx and lung). The average pixel-wise difference between the ground truth and the predicted results across all testing cases (both internal and external) is within 2%. The average 3D gamma passing rate is above 95% for all testing cases.
CONCLUSIONS: We have developed a novel deep learning framework that effectively maps radiation fluences to dose distributions. The high accuracy and efficiency of the proposed approach indicate its potential for use in online adaptive plan optimization.
METHOD: To improve the effectiveness of the dose calculation, deep learning methods are widely adopted to compute dose distributions from several variables mimicking the physical features required for dose calculation. However, these methods employed similar concepts and neural network inputs, and had only limited improvements to traditional dose calculation methods. In this work, we propose a new computing process, named DeepBEVdose, which is essentially distinct to the previous deep learning-based dose calculation methods. We present a novel image-domain dose calculation algorithm to automatically compute dose distributions from the computer tomography images and radiation field fluence maps. Specifically, a novel beam's eye view calculation scheme is introduced to substitute the traditional trivial ray-tracing procedure that cannot be removed in the previously published dose calculation system. Under this new calculation scheme, a generic two-dimensional convolutional neural network with minimal input requirements is sufficient to perform accurate dose calculation.
RESULTS: We demonstrate the feasibility of this approach with datasets acquired from multi-institutions with two different tumor sites (nasopharynx and lung). The average pixel-wise difference between the ground truth and the predicted results across all testing cases (both internal and external) is within 2%. The average 3D gamma passing rate is above 95% for all testing cases.
CONCLUSIONS: We have developed a novel deep learning framework that effectively maps radiation fluences to dose distributions. The high accuracy and efficiency of the proposed approach indicate its potential for use in online adaptive plan optimization.