文献速递头颈部肿瘤放疗放疗物理与技术

深度学习剂量组学预测鼻咽癌放疗后放射性皮炎

Deep learning dosiomics for the pretreatment prediction of radiation dermatitis in nasopharyngeal carcinoma patients treated with radiotherapy.
PubMed 文献 · Radiother Oncol · 2025-05-22

本研究基于两个医疗中心290例接受放疗的鼻咽癌患者,回顾性构建并验证融合剂量组学特征、卷积神经网络深度学习特征与临床因素的模型,用于放疗前预测≥2级放射性皮炎。联合模型在内部和外部验证集的AUC分别为0.863和0.832,整合临床因素的列线图外部验证AUC为0.832。

要点

  • 深度学习模型ResNet-34预测≥2级放射性皮炎的内部和外部验证AUC分别为0.820和0.812。
  • 融合深度学习与剂量组学特征后内部和外部验证AUC提升至0.863和0.832。
  • 整合深度学习、剂量组学与临床因素的列线图训练集AUC达0.945,外部验证为0.832。

核心结论

  • 融合深度学习、剂量组学与临床特征可有效预测鼻咽癌放疗患者的≥2级放射性皮炎,有助于改善患者管理。

研究设计与样本

  • 回顾性研究,纳入两个医疗中心290例接受放疗的鼻咽癌患者,分为训练集167例、内部验证集72例和外部验证集51例。
  • 从放疗剂量分布中提取剂量组学特征与卷积神经网络深度学习特征,构建端到端模型,并以ROC曲线AUC评估比较模型效能。

主要发现

  • 最终筛选25个剂量组学特征的XGBoost模型在内部和外部验证集的AUC分别为0.751和0.746。
  • ResNet-34深度学习模型在内部和外部验证集的AUC分别为0.820和0.812。
  • 融合深度学习与剂量组学特征后内部和外部验证集AUC分别为0.863和0.832。
  • 整合深度学习、剂量组学特征与临床因素的列线图在训练集、内部验证集和外部验证集的AUC分别为0.945、0.916和0.832。

临床怎么用

  • 可基于放疗剂量分布构建的联合模型在放疗前识别≥2级放射性皮炎高风险鼻咽癌患者,辅助个体化管理。

局限性

  • 本研究为回顾性研究,样本来自两个医疗中心,外部验证集仅51例,结论需更大样本前瞻性验证。

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

原文

PURPOSE: To develop a combined dosiomics and deep learning (DL) model for predicting radiation dermatitis (RD) of grade ≥ 2 in patients with nasopharyngeal carcinoma (NPC) after radiation therapy (RT) based on radiation dose distribution. MATERIALS AND METHODS: A retrospective study was performed with 290 NPC patients treated with RT from two medical centers. The patients were categorized into three groups: a training set (n = 167), an internal validation set (n = 72), and an external validation set (n = 51), respectively. Dosiomic features, in conjunction with DL features derived from convolutional neural networks, were extracted and analyzed from the radiation dose distribution to construct an end-to-end model and facilitate the prediction of RD. The efficacy of the developed models was assessed and compared using the area under curve (AUC) of the receiver operating characteristic (ROC) curves. RESULTS: The XGBoost model with finally screened 25 dosiomic features achieved the best AUC of 0.751 and 0.746 in the internal and external validation sets, respectively. DL model with ResNet-34 achieved the best AUC of 0.820 and 0.812 in the internal and external validation sets, respectively. Combining DL and dosiomic features improved the AUC to 0.863 and 0.832 in the internal and external validation sets, respectively. Nomogram integrating DL, dosiomic features, and clinical factors achieved an AUC of 0.945, 0.916, and 0.832 in the training, internal, and external validation sets, respectively. CONCLUSION: The integration of DL, dosiomics and clinical features is feasible and effective for predicting RD, thereby enhancing the management of NPC patients treated with RT.

来源:Radiother Oncol,查看原文。版权归原作者所有。

引用

  1. Xiang Z, Yu X, Lin S, et al. Deep learning dosiomics for the pretreatment prediction of radiation dermatitis in nasopharyngeal carcinoma patients treated with radiotherapy. Radiother Oncol. 2025. PMID: 40412532. doi:10.1016/j.radonc.2025.110951

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。