Journal of applied clinical medical physics · 2026
An interpretable radiomics-deep learning nomogram from whole-body bone scintigraphy for MDP-avid bone metastasis prediction in NSCLC.
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.
Zhai Weihao, Li Xiaolin, Zhou Qian, Zhou Taohu, Wang Yi, Zhou Xiuxiu, Jiang Xin'ang, Zhang Ziwei, Jin Qianxi, Liu Shiyuan, Fan Li
Аннотация
BACKGROUND: Whole-body bone scintigraphy (WBS) remains a widely used first-line screening tool for bone metastasis, but differentiating MDP-avid metastatic lesions from benign bone abnormalities in non-small cell lung cancer (NSCLC) remains challenging.
PURPOSE: To develop and externally validate a radiomics and deep learning bone signature (RDB) derived from planar WBS images for prediction of MDP-avid bone metastasis and to explore its prognostic value in NSCLC.
METHODS: A retrospective analysis was conducted on NSCLC patients who underwent WBS imaging across two centers. Eight hundred twenty-nine patients from Center 1 were used for training and internal validation, whereas 574 patients from Center 2 were used for external validation. Radiomics features from 2 segmentation methods and deep learning features were extracted from planar WBS images. The RDB nomogram was generated from clinical, radiomics, and deep learning features using the best-performing machine learning algorithm. Model performance was assessed using the area under the receiver operating characteristic curve. Shapley additive explanations were used to interpret the model output, and Kaplan-Meier survival analysis evaluated prognostic stratification. The locked model was also translated into a research-use-only local desktop application.
RESULTS: There were 720 patients with bone metastasis (202 solitary and 518 multiple) and 683 without bone metastasis. The final RDB nomogram included 8 radiomics features, 11 deep learning features, and 2 clinical factors. The RDB nomogram achieved an area under the curve of 0.857 (95% CI, 0.811-0.904) in the internal validation cohort and 0.870 (95% CI, 0.840-0.900) in the external validation cohort for predicting MDP-avid bone metastasis. The Rad-score was the most influential predictor. Kaplan-Meier analysis showed significant survival differences for solitary-lesion patients in both centers. A deployable desktop implementation of the final model was completed to support reproducibility and external use.
CONCLUSIONS: The RDB nomogram is a promising noninvasive tool for predicting MDP-avid bone metastasis in NSCLC patients and may provide prognostic value, particularly for solitary lesions.
PURPOSE: To develop and externally validate a radiomics and deep learning bone signature (RDB) derived from planar WBS images for prediction of MDP-avid bone metastasis and to explore its prognostic value in NSCLC.
METHODS: A retrospective analysis was conducted on NSCLC patients who underwent WBS imaging across two centers. Eight hundred twenty-nine patients from Center 1 were used for training and internal validation, whereas 574 patients from Center 2 were used for external validation. Radiomics features from 2 segmentation methods and deep learning features were extracted from planar WBS images. The RDB nomogram was generated from clinical, radiomics, and deep learning features using the best-performing machine learning algorithm. Model performance was assessed using the area under the receiver operating characteristic curve. Shapley additive explanations were used to interpret the model output, and Kaplan-Meier survival analysis evaluated prognostic stratification. The locked model was also translated into a research-use-only local desktop application.
RESULTS: There were 720 patients with bone metastasis (202 solitary and 518 multiple) and 683 without bone metastasis. The final RDB nomogram included 8 radiomics features, 11 deep learning features, and 2 clinical factors. The RDB nomogram achieved an area under the curve of 0.857 (95% CI, 0.811-0.904) in the internal validation cohort and 0.870 (95% CI, 0.840-0.900) in the external validation cohort for predicting MDP-avid bone metastasis. The Rad-score was the most influential predictor. Kaplan-Meier analysis showed significant survival differences for solitary-lesion patients in both centers. A deployable desktop implementation of the final model was completed to support reproducibility and external use.
CONCLUSIONS: The RDB nomogram is a promising noninvasive tool for predicting MDP-avid bone metastasis in NSCLC patients and may provide prognostic value, particularly for solitary lesions.