基于生化指标与CT灌注参数的机器学习模型在早期预测重症急性胰腺炎发生中的临床价值

Clinical Value of Machine Learning Model Based on Biochemical Indexes and CT Perfusion Parameters in Early Prediction of Severe Acute Pancreatitis

目的:基于生化指标和CT灌注参数,探究机器学习方法早期预测重症急性胰腺炎(AP)发生的临床价值。方法:收集本院与德阳市人民医院临床确诊AP的初诊患者121例。依据病情严重程度,分为非重症组(包括35例轻症与36例中重症)和重症组(50例重症);再按数据类型(生化指标、灌注数据以及包含生化指标和CT灌注参数的联合数据)分别进行机器学习处理。结果:生化指标模型测试集准确率及曲线下面积(AUC)值分别为0.741±0.010、0.749±0.019;灌注数据模型测试集准确率及AUC值分别为0.627±0.010、0.622±0.028;联合数据模型测试集准确率及AUC值分别为0.751±0.009、0.796±0.021。统计学验证灌注数据构建的模型的AUC值与其余2个队列均有显著差异(P<0.05),联合数据构建模型AUC值与生化指标模型AUC值无显著差异(P>0.05)。结论:本研究基于各类型数据构建的所有模型均能在疾病早期实现对重症急性胰腺炎发生的预测,且基于联合数据与生化指标构建的模型整体较灌注数据模型表现出更高的预测价值。

Purpose:To explore the clinical value of machine learning method in early prediction of severe acute pancreatitis(AP)based on biochemical indexes and CT perfusion parameters.Methods:One hundred and twenty-one patients with AP were collected from the Affiliated Hospital of North Sichuan Medical College and Deyang People’s Hospital.According to the severity of AP,patients were divided into non-severe group(including 35 mild AP cases and 36 moderate-severe AP cases)and severe group(50 severe AP cases);then machine learning was performed according to the data types(biochemical indexes,perfusion data and joint data containing biochemical indexes and CT perfusion parameters).Results:The accuracy and area under the curve(AUC)of biochemical index model test set were 0.741±0.010 and 0.749±0.019.The accuracy and AUC of perfusion data model test set were 0.627±0.010 and0.622±0.028.The accuracy and AUC of the combined data model test set were 0.751±0.009 and 0.796±0.021.The AUC value of the perfusion data model was significantly different from the other two models(P<0.05).There was no significant difference in AUC value between the joint data model and the biochemical indicator model(P>0.05).Conclusion:All models constructed based on various types of data in this study can predict the occurrence of severe AP in the early stage of disease,and the model constructed based on joint data and biochemical indicators shows higher predictive value than the perfusion data model.