基于人工神经网络的眼镜蛇咬伤辅助诊断研究

Research on auxiliary diagnosis of cobra bite based on artificial neural network

目的:探讨人工神经网络运用于眼镜蛇咬伤辅助诊断的可行性。方法:选取2017—2019年梧州市中医医院蛇伤科专科医生明确诊断的101例眼镜蛇咬伤住院患者的病例资料,并选取同期101例其他毒蛇咬伤住院患者的病例资料。根据2018年中国蛇伤救治专家共识,结合梧州市中医医院眼镜蛇咬伤诊疗方案,提取病例中12个局部症状特征和10个全身症状特征数据。首先,将特征数据进行规范化处理,建立数据集,并将其划分为训练集和测试集;其次,构建概率神经网络(probabilistic neural network,PNN)、径向基函数(radial basis function,RBF)神经网络和反向传播(back propagation,BP)神经网络进行模型训练和测试;最后采用预测准确率对模型进行性能评价。结果:PNN模型预测准确率为87.14%,RBF神经网络模型预测准确率为82.67%,BP神经网络模型预测准确率为86.02%,PNN模型预测眼镜蛇咬伤的性能优于RBF神经网络模型和BP神经网络模型。结论:基于蛇伤患者局部和全身症状特征,利用人工神经网络构建眼镜蛇咬伤辅助诊断模型对眼镜蛇咬伤进行识别理论上是可行的。其中,PNN模型预测准确率最高、泛化能力最好,更适用于眼镜蛇咬伤的辅助诊断。

Objective To discuss the feasibility of using artificial neural network to assist the diagnosis of cobra bite.Methods The data of 101 inpatients with cobra bites diagnosed by the specialist of the snakebite department in Wuzhou Hospital of Traditional Chinese Medicine from 2017 to 2019 were selected,along with the data of another 101 inpatients with other venomous snakebites in the same period.The feature on 12 local symptoms and 10 systemic ones were extracted from the cases selected based on 2018 Expert Consensus on China Snake-bites Rescue and Treatment and the treatment protocol for cobra bites in Wuzhou Hospital of Traditional Chinese Medicine.Firstly,the feature data were normalized and enrolled into a dataset,which was divided into a training set and a test set;secondly,probabilistic neural network(PNN),radial basis function(RBF)neural network and back propagation(BP)neural network were constructed for model training and testing;finally,prediction accuracy was used to evaluate the performances of the models established.Results The prediction accuracy was 87.14%by PNN model,82.67%by RBF neural network neural network model and 86.02%by BP neural network model.PNN neural network model behaved slightly better than RBF and BP neural network models.Conclusion It is theoretically feasible to construct an auxiliary diagnosis model to identify cobra bites based on the local and systemic symptoms of snakebite patients and artificial neural networks.PNN model gains advantages in prediction accuracy and generalization ability over others,and is more suitable for the auxiliary diagnosis of cobra bites.