Journal of applied clinical medical physics · 2026
A hybrid deep learning framework for automated treatment planning in cervical cancer radiotherapy.
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.
Jiang Haiyan, Yuan Zengtai, Liu Zihao, Wang Yuxiang, Hu Wei, Yan Bing, Yang Yidong
Аннотация
BACKGROUND: Radiation therapy treatment planning is a time-consuming trial-and-error process, and the plan quality is heavily dependent on planners' experiences, resulting in a strong demand for automated planning methods that can rapidly generate uniformly high-quality plans.
PURPOSE: This study aimed to develop a hybrid automated deep learning-based plan optimization (HALO) framework, integrating deep learning fluence prediction and optimization, multileaf collimator (MLC) sequencing, and GPU-accelerated Monte Carlo dose computation. HALO was designed to ensure robust plan generation for cervical cancer intensity-modulated radiation therapy (IMRT).
METHODS: The proposed HALO framework incorporated several functional modules. First, a dose-guided Fluence Prediction Network (DG-FPN) was developed. A total of 120 cervical cancer clinical IMRT plans were collected to train the DG-FPN, of which 90 plans were assigned for training, 10 for validation, and 20 for testing. The neural network took patients' computed tomography (CT) anatomy as input and predicted the 3D dose distribution and fluence maps. Next, the predicted fluence maps were further optimized to meet dose-volume constraints (DVCs), and then the refined fluence maps were converted into deliverable segments using an MLC sequencing algorithm. Finally, dose calculation was performed using a GPU-accelerated Monte Carlo engine. The 20 patients in the testing set were used to evaluate the HALO method, and the plan deliverability was validated by patient-specific IMRT Quality Assurance (QA).
RESULTS: The DG-FPN achieved superior fluence prediction accuracy compared to previous work, with a median mean absolute error (MAE) of 0.055 and a structural similarity index (SSIM) of 0.94. The automated framework generated high-quality plans and reduced dose to adjacent OARs, with V50Gy decreased from 46.5 ± 5.0% to 41.5 ± 4.8% (p = 0.008)for bladder, V35Gy decreased from 31.5 ± 8.2% to 29.2 ± 7.5% (p = 0.024) for small intestine, while preserving PTV homogeneity and conformity. Importantly, the entire treatment planning time was within 3 minutes, with fluence optimization time decreased by an average of 82.6% after using deep learning prediction. The mean gamma passing rate under the 2%/2 mm criterion for the patient-specific IMRT QA achieved 97.37 ± 1.01%.
CONCLUSIONS: This study demonstrated the clinical feasibility of the proposed HALO framework for cervical cancer radiotherapy in producing high-quality and deliverable IMRT plans. The proposed automated paradigm can serve as a stand-alone platform for treatment planning.
PURPOSE: This study aimed to develop a hybrid automated deep learning-based plan optimization (HALO) framework, integrating deep learning fluence prediction and optimization, multileaf collimator (MLC) sequencing, and GPU-accelerated Monte Carlo dose computation. HALO was designed to ensure robust plan generation for cervical cancer intensity-modulated radiation therapy (IMRT).
METHODS: The proposed HALO framework incorporated several functional modules. First, a dose-guided Fluence Prediction Network (DG-FPN) was developed. A total of 120 cervical cancer clinical IMRT plans were collected to train the DG-FPN, of which 90 plans were assigned for training, 10 for validation, and 20 for testing. The neural network took patients' computed tomography (CT) anatomy as input and predicted the 3D dose distribution and fluence maps. Next, the predicted fluence maps were further optimized to meet dose-volume constraints (DVCs), and then the refined fluence maps were converted into deliverable segments using an MLC sequencing algorithm. Finally, dose calculation was performed using a GPU-accelerated Monte Carlo engine. The 20 patients in the testing set were used to evaluate the HALO method, and the plan deliverability was validated by patient-specific IMRT Quality Assurance (QA).
RESULTS: The DG-FPN achieved superior fluence prediction accuracy compared to previous work, with a median mean absolute error (MAE) of 0.055 and a structural similarity index (SSIM) of 0.94. The automated framework generated high-quality plans and reduced dose to adjacent OARs, with V50Gy decreased from 46.5 ± 5.0% to 41.5 ± 4.8% (p = 0.008)for bladder, V35Gy decreased from 31.5 ± 8.2% to 29.2 ± 7.5% (p = 0.024) for small intestine, while preserving PTV homogeneity and conformity. Importantly, the entire treatment planning time was within 3 minutes, with fluence optimization time decreased by an average of 82.6% after using deep learning prediction. The mean gamma passing rate under the 2%/2 mm criterion for the patient-specific IMRT QA achieved 97.37 ± 1.01%.
CONCLUSIONS: This study demonstrated the clinical feasibility of the proposed HALO framework for cervical cancer radiotherapy in producing high-quality and deliverable IMRT plans. The proposed automated paradigm can serve as a stand-alone platform for treatment planning.