文献速递泌尿系结石

三维纹理分析联合机器学习可提升肾结石体外冲击波碎石成功预测

Three-Dimensional Texture Analysis with Machine Learning Provides Incremental Predictive Information for Successful Shock Wave Lithotripsy in Patients with Kidney Stones.
PubMed 文献 · J Urol · 2018-04-17

本研究纳入结石直径5至20 mm、既往未治疗且有术前术后CT的肾结石患者,提取224个三维纹理特征,用5种机器学习模型建模,按2/3推导、1/3验证划分数据。验证队列中,身体质量指数、初始结石大小和皮肤至结石距离的AUC分别为0.68、0.58和0.63,HU无预测价值;随机森林用3个三维纹理特征AUC为0.79,联合皮肤至结石距离后AUC达0.85。

要点

  • 身体质量指数、初始结石大小和皮肤至结石距离对碎石成功的预测价值有限。
  • 结石HU值在本研究中未显示出对碎石成功的预测价值。
  • 机器学习筛选的3个三维纹理特征可提升碎石成功预测准确性。

核心结论

  • 临床变量身体质量指数、初始结石大小和皮肤至结石距离对预测碎石成功价值有限,结石HU值无预测价值。
  • 机器学习筛选的选定三维纹理特征可为预测碎石成功提供增量准确性。

研究设计与样本

  • 纳入有术前和术后CT、既往未治疗、结石直径5至20 mm的肾结石患者。
  • 分析每个结石的224个三维纹理特征,包括HU测定的衰减值及身体质量指数、初始结石大小和皮肤至结石距离等临床变量,使用5种常用机器学习模型。
  • 数据集按2/3用于模型推导、1/3用于验证划分,验证队列通过计算敏感度、特异度和AUC评估机器学习预测碎石成功。

主要发现

  • 验证队列中,身体质量指数、初始结石大小和皮肤至结石距离预测碎石成功的AUC分别为0.68、0.58和0.63。
  • HU未发现预测价值。
  • 使用3个三维纹理分析特征的随机森林分类器AUC为0.79。
  • 3个纹理特征联合皮肤至结石距离AUC为0.85,联合身体质量指数AUC为0.8,联合结石大小AUC为0.81。

临床怎么用

  • 对于直径5至20 mm、既往未治疗的肾结石患者,可考虑基于CT三维纹理分析联合机器学习预测体外冲击波碎石成功概率。
  • 身体质量指数、初始结石大小和皮肤至结石距离可作为有限参考,HU值不宜单独用于预测碎石成功。

局限性

  • 本研究为初步研究。

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: We sought to determine the predictive value of 3-dimensional texture analysis of computerized tomography images for successful shock wave lithotripsy in patients with kidney stones. MATERIALS AND METHODS: Patients with preoperative and postoperative computerized tomography, previously untreated kidney stones and a stone diameter of 5 to 20 mm were included in study. A total of 224, 3-dimensional texture analysis features of each kidney stone, including attenuation measured in HU and the clinical variables body mass index, initial stone size and skin to stone distance, were analyzed using 5 commonly used machine learning models. The data set was split in a ratio of 2/3 for model derivation and 1/3 for validation. Machine learning based predictions of shock wave lithotripsy success in the validation cohort were evaluated by calculating sensitivity, specificity and the AUC. RESULTS: For shock wave lithotripsy success the 3 clinical variables body mass index, initial stone size and skin to stone distance showed an AUC of 0.68, 0.58 and 0.63, respectively. No predictive value was found for HU. A random forest classifier using 3, 3-dimensional texture analysis features had an AUC of 0.79. By combining these 3 features with clinical variables discriminatory accuracy improved further with an AUC of 0.85 for 3-dimensional texture analysis features and skin to stone distance, an AUC of 0.8 for 3-dimensional texture analysis features and body mass index, and an AUC of 0.81 for 3-dimensional texture analysis and stone size. CONCLUSIONS: This preliminary study indicates that the clinical variables body mass index, initial stone size and skin to stone distance show limited value to predict shock wave lithotripsy success while stone HU values were not predictive. Select 3-dimensional texture analysis features identified by machine learning provided incremental accuracy to predict the success of shock wave lithotripsy.

来源:J Urol,查看原文。版权归原作者所有。

引用

  1. Mannil M, von Spiczak J, Hermanns T, et al. Three-Dimensional Texture Analysis with Machine Learning Provides Incremental Predictive Information for Successful Shock Wave Lithotripsy in Patients with Kidney Stones. J Urol. 2018. PMID: 29673945. doi:10.1016/j.juro.2018.04.059

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。