[Paper Review] Medical Text Classification using Convolutional Neural Networks

Mark Hughes, Irene Li+2 more|PubMed|Apr 22, 2017

Topic Modeling152 citations

TL;DR

The paper uses deep convolutional neural networks to classify clinical text at the sentence level and reports a ~15% improvement over common NLP baselines.

ABSTRACT

We present an approach to automatically classify clinical text at a sentence level. We are using deep convolutional neural networks to represent complex features. We train the network on a dataset providing a broad categorization of health information. Through a detailed evaluation, we demonstrate that our method outperforms several approaches widely used in natural language processing tasks by about 15%.

Motivation & Objective

• Motivate automatic sentence-level classification of clinical text.
• Explore representation of complex features via convolutional neural networks.
• Evaluate CNN-based approach on a dataset with broad health information categories.
• Compare CNN performance against widely used NLP baselines.

Proposed method

• Represent clinical sentences with deep convolutional neural networks to capture complex features.
• Train CNN model on a dataset providing broad health information categories.
• Evaluate performance against standard NLP approaches widely used in the field.
• Quantitatively assess improvement over baselines (reported as about 15%).

Experimental results

Research questions

• RQ1 Can CNNs effectively classify clinical text at the sentence level?
• RQ2 Do CNN-based representations outperform traditional NLP approaches on medical text classification?
• RQ3 What is the magnitude of improvement achieved by CNNs on the given dataset?

Key findings

• CNN-based approach outperforms several widely used NLP methods by about 15%.
• The method demonstrates effective representation of complex features in clinical text.
• The evaluation validates the CNN model on a dataset with broad health information categories.