Abstract: Objective Based on the artificial neural network diagram,a sub-model of failure factors of chronic subdural hematoma treated by skull drilling and drainage was constructed,in order to provide clinical basis for diagnosis and treatment.Methods A total of 300 CSDH patients admitted to our hospital from April 2020 to August 2021 were selected as the research subjects.After hospitalization,all patients were treated with cranial trephination and drainage.According to the treatment results,they were divided into successful group (n=252) and failed group (n=48).The clinical data of the two groups of patients were compared,and the factors affecting treatment failure of CSDH patients were analyzed by multivariate Logistic regression method.The artificial neural network model was constructed,and the author characteristic (ROC) curve was used to evaluate the differentiation of the model and the effectiveness of the clinical decision curve.Results The age,operation time,intraoperative blood loss,serum NSE,proportion of secondary intracranial hematoma,serum TSP1 and TSP2 in the successful group were significantly lower than those in the failed group (P＜0.05),and the average diameter of bone pore was significantly higher than that in the failed group (P＜0.05).Multivariate analysis showed that long operation time,large amount of intraoperative blood loss,increased serum NSE level,secondary intracranial hematoma,and increased TSP1 and TSP2 levels were independent risk factors for the treatment failure of CSDH patients by cranial trephine drainage (P＜0.05).However,slightly larger mean bone pore diameter was a protective factor for CSDH patients' treatment failure through cranial drainage (P＜0.05).ROC curve showed that the AUC of the artificial neural network model was 0.935(95%CI:0.871~0.986,P＜0.001),and clinical decision curve showed that the artificial neural network model was effective.Conclusion Clinical observation of CSDH patients' operation time,intraoperative blood loss,serum NSE leveletc,can predict the therapeutic effect of cranial drilling and drainage and take corresponding prevention and treatment measures to reduce the mortality of CSDH patients.

Key words: Cranial drilling and drainage, Chronic subdural hematoma, Therapeutic effect, Clinical decision curve, Artificial neural network model

摘要： 目的 基于人工神经网络图构建慢性硬膜下血肿行颅骨钻孔引流术治疗失败的因素分模型,以期为临床提供诊疗依据。方法 选取2020年4月至2021年8月收治于我院的300名CSDH患者作为研究对象,住院后全部采用颅骨钻孔引流术治疗,根据治疗结果分为成功组(n=252)和失败组(n=48)。比较两组患者的临床资料,采用多因素Logistic回归法分析影响CSDH患者治疗失败的因素,并构建人工神经网络模型,受试工作者特征(ROC)曲线评价模型的区分度,临床决策曲线评价模型的有效性。结果 成功组的年龄、手术时间、术中出血量、血清NSE、继发性颅内血肿比例、血清TSP1和TSP2均明显低于失败组(P＜0.05),平均骨孔直径明显高于失败组(P＜0.05)。多因素分析结果显示,手术时间长、术中出血量多、血清NSE水平升高、继发性颅内血肿、TSP1和TSP2水平升高是CSDH患者经颅骨钻孔引流术治疗失败的独立危险因素(P＜0.05),而平均骨孔直径略大是CSDH患者经颅骨钻孔引流术治疗失败的保护因素(P＜0.05)。ROC曲线显示人工神经网络模型的AUC为0.935(95%CI:0.871~0.986,P＜0.001),临床决策曲线显示人工神经网络模型的有效性较好。结论 临床可以通过观察CSDH患者的手术时间、术中出血量、血清NSE水平等来预测颅骨钻孔引流术的治疗效果并采取对应的防治措施,降低CSDH患者的死亡率。

关键词: 颅骨钻孔引流术, 慢性硬膜下血肿, 治疗效果, 临床决策曲线, 人工神经网络模型