Review of Natural Language Processing in Radiology

https://doi.org/10.1016/j.nic.2020.08.001Get rights and content

Section snippets

Key points

  • Natural language processing (NLP) will enable a variety of assistive radiologic applications.

  • Advances building on NLP techniques will permit machines to understand, classify, summarize, and generate text to automate linguistic tasks.

  • NLP algorithms can be complementary to other radiology imaging artificial intelligence applications for diagnostic or workflow enhancement.

What is natural language processing?

NLP is defined as a branch of artificial intelligence that is concerned with the interaction between computers and humans using natural language. NLP is a multidisciplinary field that combines classic linguistics with traditional computer science and modern artificial intelligence (AI) methods. The intention of NLP is to enable machines to read and understand human languages for meaningful purposes. Given the diversity of tasks possible, there are potentially many different customizations or

Definition

Although simple tokenization of words allows the mapping of each individual word to an index as a one-dimensional array or vector, those tokenization approaches miss the context 2 similar words might have with one another. For instance, a model that assigns a numerical index to a word based on alphabetical position might place “king” and “queen” or “men” and “women” very far apart, but these terms have closely related higher-level concepts (ie, “royal titles”, “gender”).

Instead of simply

Regular Expressions

Regular expressions, additionally termed as regex or regexp is a sequence of characters that define a search pattern, often used for searching and matching patterns found in strings. Although more commonly known whole search keyword matching is encompassed within the functionality of regular expressions, the concept and syntax of regular expressions is an extremely compact but expressive manner in which to compose matching rules for strings. Originating from theoretic works on regular languages

Automatic Protocoling

Some investigators have proposed using machine learning to automate magnetic resonance (MR) imaging protocol selection of radiology requisitions. In a study by Brown and Marotta,39 a machine learning model was developed to classify unstructured clinical history indications and assign MR imaging protocols, with attempted models comparing support vector machine, gradient boosting, and random forest techniques. The most performant model, a gradient boosting machine, was able to achieve a protocol

Limitations

Working with NLP poses different challenges and obstacles in comparison with machine learning in computer vision, some of which are unique to linguistics. One key difficulty when working with NLP problems concerns the accessibility of training datasets. In contrast with the longer open-science public record of recent machine learning and AI competitions and imaging datasets, because of the inherently private nature of reports and biomedical text, large public datasets of medical records

Summary

The ability to generalize with machine learning and AI technologies permits a wide gamut of possibilities and applications. Although much attention has been paid to image interpretative tasks, numerous other opportunities exist with NLP that offer similar or greater clinical value, often as an adjunctive or assistive technology, which could lead to improved clinical workflows, greater safety and efficiencies, and improved patient quality of life and health care satisfaction. Rather than posing

Disclosure

J.W. Luo and J.J.R. Chong: no relevant disclosures.

First page preview

First page preview
Click to open first page preview

References (51)

  • G. Chartrand et al.

    Deep learning: a primer for radiologists

    Radiographics

    (2017)
  • Webster JJ, Kit C. Tokenization as the initial phase in NLP. In: COLING 1992 Volume 4: The 15th International...
  • Silva C, Ribeiro B. The importance of stop word removal on recall values in text categorization. In: Proceedings of the...
  • Balakrishnan V, Lloyd-Yemoh E. Stemming and lemmatization: a comparison of retrieval performances. In: Proceedings of...
  • Plisson J, Lavrac N, Mladenic D. A rule based approach to word lemmatization. Proceedings of IS-2004. Salt Lake City...
  • Brill E. A simple rule-based part of speech tagger. In Proceedings of the third conference on Applied natural language...
  • Navigli R. Word sense disambiguation: A survey. ACM computing surveys (CSUR)...
  • Y. Zhang et al.

    Understanding bag-of-words model: a statistical framework

    International Journal of Machine Learning and Cybernetics

    (2010)
  • E.M. Voorhees

    Natural language processing and information retrieval

  • Ramos J. Using tf-idf to determine word relevance in document queries. In Proceedings of the first instructional...
  • Taylor SJ, Harabagiu SM. The Role of a Deep-Learning Method for Negation Detection in Patient Cohort Identification...
  • Mikolov T, Sutskever I, Chen K, et al. Distributed representations of words and phrases and their compositionality. In...
  • Pennington J, Socher R, Manning C. Glove: Global vectors for word representation. In Proceedings of the 2014 conference...
  • Banerjee I, Madhavan S, Goldman RE, et al. Intelligent word embeddings of free-text radiology reports. In AMIA Annual...
  • S.C. Kleene

    Representation of events in nerve nets and finite automata

    (1951)
  • Cited by (31)

    • Artificial intelligence and personalized medicine: transforming patient care

      2023, The New Era of Precision Medicine: What it Means for Patients and the Future of Healthcare
    • Applications of natural language processing in radiology: A systematic review

      2022, International Journal of Medical Informatics
      Citation Excerpt :

      NLP approaches were grouped into three broad categories: rule-based methods, classical machine learning, and deep learning. The categorization was based on definitions from prior publications [6] as well as the description of methods within each individual publication. Rule-based methods apply predefined rules to manipulate words, phrases, or sentences.

    • Application of Short Video Description Technology in College English Teaching

      2024, International Journal of Web-Based Learning and Teaching Technologies
    View all citing articles on Scopus
    View full text