运用人工神经网络（ANN）技术预测颅脑损伤患者的预后效果。

在已完成的与预后相关的入院指标的二次多项式回归模型研究的基础上，回顾性分析2013年1月至2017年8月入住扬州大学附属医院重症医学科的130例颅脑损伤患者的临床资料，建立基于ANN技术的颅脑损伤患者预后预测模型，同时采用2017年10月至2019年3月46例颅脑损伤患者临床资料进行外部验证，计算其相关系数、敏感度及特异度等参数，并与二次多项式逐步回归模型进行对比分析。

ANN技术建立的颅脑损伤患者预后预测模型，内部验证中，其相关系数为0.8935，不良预后的敏感度为94.8%（55/58），特异度为82.1%（55/67）；良好预后的敏感度为95.5%（42/44），特异度为87.5%（42/48）。外部验证中，其相关系数为0.7138，不良预后的敏感度为43.8%（7/16），特异度为100.0%（7/7）；良好预后的敏感度为100.0%（26/26），特异度为66.7%（26/39）。

与二次多项回归模型比较，ANN技术建立的预测颅脑损伤患者预后的数学模型的模型拟合程度较高，但对于预后评估的敏感度及特异度，优势不明显。

To apply an artificial neural network (ANN) in patients with craniocerebral injury to characterize its prognostic ability.

Based on the completed quadratic polynomial stepwise regression analysis, data of 130 patients with brain injury from January 2013 to August 2017 in the Department of Critical Care Medicine of the Affiliated Hospital of Yangzhou University were collected. A prognostic prediction model was established based on artificial neural network technology. Then the clinical data of 46 patients with craniocerebral injury admitted from October 2017 to March 2019 were used for external verification, the correlation coefficient, sensitivity and specificity were calculated, and compared with the quadratic polynomial stepwise regression model.

Artificial neural network technology could be used to establish a prognosis prediction model for patients with craniocerebral injury. The correlation coefficient was 0.8935. In internal verification, the sensitivity of "poor outcome" (GOS score 1 or 2 points) was 94.8% (55/58), the specificity was 82.1% (55/67); the sensitivity of "good outcome" (GOS score 4 or 5 points) was 95.5% (42/44), and the specificity was 87.5% (42/48). The external verification correlation coefficient was 0.7138, the sensitivity of "poor outcome" was 43.8% (7/16), the specificity was 100.0% (7/7); the sensitivity of "good outcome" was 100.0% (26/26), the specificity was 66.7% (26/39).

The mathematical model established by artificial neural network technology has a better fitting than the quadratic polynomial stepwise regression model for predicting the prognosis of patients with brain injury. However, for the sensitivity and specificity of prognostic evaluation, the neural network model does not show obvious advantages.