CT组学特征在急性颅脑损伤患者近期预后的应用研究

Application study of CT radiomics features in short-term prognosis of patients with acute craniocerebral injury

目的:探讨CT影像组学特征对急性颅脑损伤(ACI)患者近期预后的预测作用。方法:回顾性分析2021年3月—2022年6月某院收治的100例ACI患者的临床资料,根据患者治疗3个月后的格拉斯哥昏迷评分(GCS)结果将患者分为预后良好组(n=57)和预后不良组(n=43);依据患者入院时CT影像组学参数,采用ITk-snap软件绘制颅脑损伤和血肿的感兴趣区域(ROIs),提取CT影像组学特征;使用曲线下面积(AUC)分析及五折交叉验证的最低绝对收缩和选择算子LASSO回归算法筛选特征,使用Spearman秩相关分析剔除冗余特征,并以此构建基于支持向量机(SVM)的预测模型。另前瞻性选取30例新入院的ACI患者,分别依据上述SVM模型和常规赫尔辛基CT评分量表评估预后,受试者工作特征(ROC)曲线比较两者的预测效能。结果:SVM模型分析结果显示,影响近期预后不良的前6位特征是短区间高灰度值强度、灰度级方差、灰度级非均匀性、小区域高灰度值强调,归一化尺寸区域非均匀性及大面积突出值。ROC曲线分析显示,SVM模型预测效能优于常规赫尔辛基CT评分量表,SVM模型的AUC值为0.921,最佳截断值为0.442,对应的灵敏度、特异度分别为0.884、0.860。结论:基于CT影像组学特征构建的ACI患者近期预后SVM模型具有较好的预测效能,可为临床预防ACI患者治疗后出现死亡、复发等不良预后提供依据。

Objective:To explore the predictive effect of CT radiomics features on short-term prognosis of patients with acute craniocerebral injury(ACI).Methods:Clinical data of 100 ACI patients treated in a hospital from March 2021 to June 2022 were retrospectively analyzed.The patients were divided into a good prognosis group(n=57)and a poor prognosis group(n=43)after 3 months’treatment based on the result of Glasgow coma scale(GCS).According to the CT radiomics parameters of the patients on admission,regions of interest(ROIs)of craniocerebral injury were drawn with ITk-snap and CT radiomics features were extracted.The features were screened with AUC analysis and LASSO regression of 5-fold cross validation method.Spearman correlation analysis was used to eliminate redundant features.Predictive model based upon the SVM was set up accordingly.Another 30 ACI patients were prospectively chosen on admission.Their prognosis was appraised with SVM model and Helsinki CT score and predictive effect was compared using ROC.Results:SVM model analysis indicated that the first 6 features influencing the short-term poor prognosis were intensity of short interval high grayscale value,grayscale variance,grayscale nonuniformity,small area high gray level emphasis,nonuniformity in normalized size area and extensive protruding value.ROC analysis showed that SVM model was superior to Helsinki CT score in predicting effect.For the SVM model,its AUC value was 0.921,the optimum cutoff value was 0.442,the corresponding sensitivity and specificity were respectively 0.884 and 0.860.Conclusion:Based upon CT radiomics features,the SVM model for the short-term prognosis of ACI patients possesses good predictive effect,which can provide clinical basis for the poor prognosis of death and recurrence in ACI patients after treatment.