Deep Learning for Natural Language Processing in Radiology-Fundamentals and a Systematic Review

Area of Science:

• Artificial Intelligence
• Radiology Informatics
• Medical Imaging Analysis

Background:

• Natural language processing (NLP) converts unstructured text to structured data.
• Deep learning innovations have substantially improved NLP performance.
• Radiology reports are a rich source of clinical information.

Purpose of the Study:

• To survey deep learning NLP fundamentals.
• To review current radiology-related research utilizing deep learning for NLP.
• To understand the impact of deep learning on NLP in medical imaging.

Main Methods:

• Systematic review adhering to Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines.
• Searched MEDLINE, Scopus, and Google Scholar for studies up to September 2019.
• Identified and analyzed ten relevant deep learning NLP studies in radiology published between 2018-2019.

Main Results:

• Convolutional neural networks, recurrent neural networks, long short-term memory networks, and attention networks are key deep learning models used.
• Applications include flagging diagnoses (e.g., pulmonary embolisms, fractures), labeling follow-up recommendations, and automatic imaging protocol selection.
• Deep learning NLP models demonstrate performance equal to or exceeding traditional NLP methods.

Conclusions:

• The application and research of deep learning NLP in radiology are rapidly expanding.
• Familiarity with deep learning NLP is crucial for radiologists to adapt to evolving field.
• This technology promises to enhance efficiency and accuracy in radiological practice.