颈动脉狭窄斑块放射组学及超声列线图预测无症状患者卒中风险

Radiomics and ultrasound nomogram of carotid stenosis plaque predict the risk of stroke in asymptomatic patients

目的:开发和验证一种基于放射组学和超声特征的列线图,用于预测颈动脉狭窄导致卒中的风险。方法:回顾性纳入288例确定了有无卒中发生患者的资料,随机按7∶3比例分为训练组(n=201)和验证组(n=87)。收集患者临床数据、血管超声特征及CT血管成像(CTA)斑块特征,并从颈动脉CTA图像中提取狭窄斑块的放射组学特征。使用最小绝对收缩和选择算子分析与发生卒中相关的放射组学特征并计算影像组学评分。接着,对临床危险因素、斑块特征和放射组学评分进行单因素及多因素logistic分析,分别构建CTA斑块特征模型、超声斑块特征模型以及联合模型,并验证列线图的性能。结果:CTA斑块特征中斑块长度、厚度、溃疡、软斑块,超声斑块特征中斑块处狭窄程度、长度、厚度、狭窄处峰值流速、阻力指数是发生卒中的独立危险因素;此外,CTA斑块放射组学特征中,12个特征与患者发生卒中风险相关。基于超声、CTA和放射组学评分建立的列线图显示出良好的校准和分辨力(训练组和验证组的曲线下面积分别为0.886和0.837)。校准图和临床决策曲线显示出良好的临床校准度和实用性。结论:基于颈动脉狭窄斑块的放射组学及超声列线图对颈动脉斑块狭窄患者发生卒中风险具有良好的预测性能。

Purpose:To develop and validate a radiomics and ultrasound-based nomogram for predicting the risk of stroke caused by carotid artery stenosis.Methods:A total of 288 patients with or without stroke were retrospectively included and randomly divided into a training set(n=201)and a validation set(n=87)in a 7:3 ratio.Clinical data,ultrasound features,CTA plaque characteristics,and radiomics features of stenotic plaques were collected.The radiomics features associated with stroke were identified using LASSO regression analysis and were used to construct a radiomics score(Radscore).Single-factor and multi-factor logistic regression analyses were performed using clinical risk factors,plaque features,and Radscore to construct CTA plaque feature models(CTA_features),ultrasound plaque feature models(US_eatures),and a combined model(Combined_model),and to validate the performance of the nomogram.Results:Plaque length,plaque thickness,plaque ulceration,and soft plaque were identified as independent risk factors for stroke in CTA plaque features,while the degree of stenosis,plaque length,plaque thickness,peak flow velocity at the stenosis,and resistance index at the stenosis were independent risk factors for stroke in US plaque features.Twelve radiomics features were found to be associated with the risk of stroke.The nomogram based on US features,CTA features,and Radscore showed good calibration and excellent discrimination(with AUCs of 0.886 and 0.837 in the training set and validation set,respectively).The calibration plot and clinical decision curve showed good clinical accuracy and usefulness.Conclusions:The radiomics nomogram based on carotid artery stenotic plaque has good predictive performance for the risk of stroke in patients with carotid artery stenosis.