一种基于深度神经网络的临床记录ICD自动编码方法

An automatic ICD coding method for clinical records based on deep neural network

随着国际疾病分类(international classification of diseases,ICD)编码数量的增加,基于临床记录的人工编码难度和成本大大提高,自动ICD编码技术引起了广泛的关注。提出一种基于多尺度残差图卷积网络的自动ICD编码技术,该技术采用多尺度残差网络来捕获临床文本的不同长度的文本模式,并基于图卷积神经网络抽取标签之间的层次关系,以加强自动编码能力。在真实医疗数据集MIMIC-III上的实验结果表明,该方法的P@k和Micro-F1分别为72.2%和53.9%,显著提高了预测性能。

With the increase in the number of the international classification of diseases (ICD) codes,the difficulty and cost of manual coding based on clinical records have greatly increased,and automatic ICD coding technology has attracted widespread attention.A multi-scale residual graph convolution network automatic ICD coding technology was proposed.This technology uses a multi-scale residual network to capture text patterns of different lengths of clinical text and extracts the hierarchical relationship between labels based on the graph convolutional neural network to enhance the ability of automatic coding.The experimental results on the real medical data set MIMIC-III show that the P@k and Micro-F1 of this method are 72.2% and 53.9%,respectively,which significantly improves the prediction performance.