Neural transfer learning for assigning diagnosis codes to EMRs

作者：

Highlights：

• Transfer learning using convolutional neural networks improves multi-label learning.

• Predicting MeSH terms for biomedical articles is a useful source task for EMR coding.

• Using 2 copies of source task parameters, one fixed and one tuned, helps target models.

• Using both word embeddings and convolutions from source task improves prediction.

摘要

•Transfer learning using convolutional neural networks improves multi-label learning.•Predicting MeSH terms for biomedical articles is a useful source task for EMR coding.•Using 2 copies of source task parameters, one fixed and one tuned, helps target models.•Using both word embeddings and convolutions from source task improves prediction.

论文关键词：Medical coding,Convolutional neural networks,Transfer learning,Multi-label classification

论文评审过程：Received 19 July 2018, Revised 20 December 2018, Accepted 10 April 2019, Available online 12 April 2019, Version of Record 19 April 2019.

论文官网地址：https://doi.org/10.1016/j.artmed.2019.04.002