PhD Seminar • Natural Language Processing • Transformer-based Model for ICD Coding Classification

Wednesday, October 6, 2021 9:00 am - 9:00 am EDT (GMT -04:00)

Please note: This PhD seminar will be given online.

Georgios Michalopoulos, PhD candidate
David R. Cheriton School of Computer Science

Supervisors: Professors Ian McKillop, Helen Chen

The International Classification of Diseases (ICD) system is a standardized way for classifying diseases and procedures during a healthcare encounter. Assigning the most appropriate and accurate codes is an important task in healthcare since erroneous ICD codes could seriously affect the organization’s ability to make accurate clinical and financial decisions.

Transformer-based models have achieved state-of-the-art results in multiple NLP tasks. However, they have yet to achieve state-of-the-art results in the ICD classification task as these models are generally unable to process documents that contain a large number of tokens which is usually the case with real patient notes.

In this presentation, we propose the combined usage of a Graph Convolutional Network (using the relations between the ICD codes) and a contextual embedding model that can process larger documents for the ICD classification task. Experimental results on a real-world clinical dataset demonstrate the effectiveness of our model on the ICD code classification task as it outperforms the previous state-of-the-art models.


To join this PhD seminar on Zoom, please go to https://uwaterloo.zoom.us/j/99472949508?pwd=N0lhNnNPWjFKQ00rV2ZZN1VHT0FKUT09.