利用自动脑室容积分析和影像组学预测分流管故障
研究纳入2018年1月至2020年8月接受分流手术的患者,采集分流管置入前、分流管功能正常或分流管故障时的头部CT。自动分割脑室计算容积并与参考常模数据比较,结合影像组学与临床特征训练分类器识别分流管故障。容积分析准确度为70%至73%(AUC 0.772),支持向量机模型预测分流管故障AUC为0.92、F1分数为0.848,留出测试集AUC为0.933。
要点
- 自动脑室分割结合影像组学、年龄和性别,可在参考常模数据辅助下帮助诊断脑积水和分流管故障。
- 将病理脑室容积与常模数据比较,准确度为70%至73%,AUC为0.772,取决于SD截断值。
- 支持向量机模型预测分流管故障的AUC为0.92、F1分数为0.848,留出测试集AUC为0.933。
核心结论
- 基于卷积神经网络的自动脑室分割结合影像组学分析,并联合年龄和性别,在结合参考常模数据时可辅助诊断脑积水和分流管故障。
- 该方法较手动分割更省时,减少观察者间变异,并有望通过促进脑积水/分流管故障的早期准确诊断改善患者结局。
研究设计与样本
- 研究纳入2018年1月至2020年8月间接受分流手术的患者,收集分流管置入前、分流管功能正常或分流管故障时的头部CT。
- 对CT扫描行自动脑室分割以计算脑室容积,并将每例患者的脑室容积与参考常模数据比较,判断是否处于给定SD范围内。
- 在病理数据集与正常数据集上结合临床特征行影像组学分析,训练分类器识别分流管故障患者。
主要发现
- 共收集并分割来自66例患者的145份头部CT。
- 将病理脑室容积与常模数据比较,准确度为70%至73%,取决于SD截断值,AUC为0.772。
- 在145份病理扫描与73份正常扫描上行影像组学分析,表现最佳的模型为支持向量机模型,预测分流管故障的AUC为0.92、F1分数为0.848。
- 支持向量机模型在留出测试数据集上测试,AUC为0.933。
局限性
- 原文未报告局限性。
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
BACKGROUND AND OBJECTIVES: The assessment of ventricle size is crucial in diagnosing hydrocephalus and in detecting shunt malfunctions. Current methods primarily involve 2-dimensional measurements or ratios. We evaluated the accuracy of volumetric analysis and radiomics in diagnosing hydrocephalus and shunt malfunction. METHODS: We identified patients that underwent shunt surgery between January 2018 and August 2020 and collected head CTs from patients who were imaged before the placement of their shunt, with a functional shunt, or with a shunt malfunction. We performed automated ventricle segmentation on the computed tomography (CT) scans to compute ventricle volumes. For each patient, the ventricular volume was compared against a reference normative data set to determine if the ventricular volume was within a given range of SDs. Radiomics analyses were performed on the pathological and a normal data set, combined with clinical features, and used to train classifiers to identify patients with a malfunctioning shunt. RESULTS: A total of 145 head CTs from 66 patients were collected and segmented. Comparing pathological ventricular volumes to a normative data set yielded an accuracy of 70% to 73%, depending on the SD cutoff (area under the curve [AUC] of 0.772). When radiomics analysis was performed on 145 pathological and 73 normal scans, the best performing model was a support vector machine model that predicted malfunctioning shunt with an AUC of 0.92 and F1-score of 0.848. Furthermore, the support vector machine model was tested using a held-out testing data set that achieved an AUC of 0.933. CONCLUSION: Automated ventricle segmentation using convolutional neural networks combined with radiomics analysis can be used with age and sex to assist in the diagnosis of hydrocephalus and shunt malfunctions when combined with a reference normative data set. It offers a time-efficient alternative to manual segmentation, reduces interobserver variability, and holds promise in improving patient outcomes by facilitating early and accurate diagnosis of hydrocephalus/shunt malfunction.
引用
- Kellogg RT, Vargas J, Nguyen M, et al. Prediction of Shunt Malfunction Using Automated Ventricular Volume Analysis and Radiomics. Neurosurgery. 2024. PMID: 39589123. doi:10.1227/neu.0000000000003296
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。