基于深度学习的病历质量控制系统设计

Design of a medical record quality control system based on deep learning

摘要:

医疗领域患者的主诉信息是医疗文本分类工作的关键,能为智慧医疗和信息文本归类提供有力的支持。近几年来随着深度学习的发展应用,基于传统深度学习技术的全流程病历内涵质量控制模型层出不穷,但传统模型存在很多缺点和局限性,诸如训练速度慢、精度损失、过拟合和无法处理大规模数据的问题,因此,引入改进的深度学习算法。指南指导下基于深度学习的全流程病历内涵质量控制体系实验结果为,将词向量设置成160时双向循环神经网络(Bidirectional Recurrent Neural Network,BiGRU-SA)模型效果最优,准确率为84.9% 。BiGRU-SA MODEL,精准度受向量维度的影响并不大。而改进的文本分类式前馈神经网络(Transformation-extraction-convolutional CNN,TextCNN)模型,精准度在其进行第3次和第四次迭代更新时,发生指数级增长,并在第3次迭代时,精度达到理想值,为8.3×10_-1^。随着迭代次数的增加,模型准确率呈现先增大后减小的趋势,在进行第6次迭代时模型效果最优,准确率为84.9% 。优化后的全流程病历内涵质量控制模型在变动率指标下的面积的值、准确率、F1、召回率四项指标值都有了一定的提升,以上结果能更好地解决过拟合和特征信息丢失的问题,并且实现全流程病历内涵质量的控制。

Abstract:

The main complaint information of patients in the medical field is the key to medical text classification work, which can provide strong support for smart healthcare and information text classification. In recent years, with the development and application of deep learning, there have been numerous quality control models for the entire process of medical records based on traditional deep learning techniques. However, traditional models have many shortcomings and limitations, such as slow training speed, accuracy loss, overfitting, and inability to handle large-scale data. Therefore, improved deep learning algorithms have been introduced. The experiment result of the whole process medical record connotation quality control system based on in-depth learning under the guidance of the guide is that when the word vector is set to 160, the Bidirectional recurrent neural networks (BiGRU-SA) model has the best effect, with an accuracy rate of 84.9%. The accuracy of BiGRU-SA Model is not significantly affected by the vector dimension. However, the accuracy of the improved transformation extraction evolutionary CNN (TextCNN) model increases exponentially when it is updated in the third and fourth iterations, and reaches the ideal value of 8.3 in the third iteration × 10-1. As the number of iterations increases, the accuracy of the model shows a trend of first increasing and then decreasing. In the sixth iteration, the model performs best with an accuracy of 84.9%. The optimized whole process medical record connotation quality control model has improved the area value, accuracy, F1, and recall rate under the rate of change index. The above results can better solve the problems of overfitting and feature information loss, and achieve the control of the connotation quality of the entire process medical record.