A Unified Review of Deep Learning for Automated Medical Coding
Published in Preprint, 2022
Recommended citation: Ji, Shaoxiong, et al. "A Unified Review of Deep Learning for Automated Medical Coding." arXiv preprint arXiv:2201.02797 (2022). https://arxiv.org/abs/2201.02797
We reviews automated medical coding from an exciting perspective that unifies existing deep learning-based models into an encoder decoder framework. Specifically, we discuss 1) neural encoders with recurrent and convolutional networks and neural attention mechanism and hierarchical encoders typically used for long clinical notes; 2) mechanisms to build deep architectures, including simple stacking, embedding injection, and residual connection; 3) decoder modules with linear layers, neural attention, hierarchical and multitask decoders; 4) the usage of auxiliary information such as Wikipedia articles, code description, and code hierarchy. Besides, we introduce data for medical coding, the evaluation of medical coding models, and real-world practice. We summarize the limitation and point out future research directions at the end of this review.
Recommended citation: Ji, Shaoxiong, et al. “A Unified Review of Deep Learning for Automated Medical Coding.” arXiv preprint arXiv:2201.02797 (2022).