A Multi-channel Convolutional Neural Network for ICD Coding
Jie Chen, Fei Teng, Zheng Ma, Li Chen, Lufei Huang, Xuan Li · 2019
The processing of textual medical data is difficult because they are structurally free, diverse in style, and have subjective factors. Medical records are primarily designed for archiving patient clinical information and administrative healthcare tasks. They are encoded following the International Classification of Diseases (ICD) standard by medical informaticians. Such manual encoding task is error-prone because of a load of abbreviations, miswriting, and medical terms in medical records. In this paper, a multi-channel convolutional attentional network is presented to automatically predict ICD codes from clinic records. The multi-channel CNN generates the multiple representations of medical records and meets the needs of a huge number of labels. An attention mechanism is designed to address the correlation between the multiple medical records representations and corresponding codes. The results show that F1 score increases by 6.9% over the state of art. This coding task eases the burden of hospital manual coding task and improve the secondary use for clinical informatics.