Lothian Ultrasound Study - LingUiST

  • Research type

    Research Study

  • Full title

    The clinical utility of natural language processing and machine learning to identify common biliary pathology from radiology reports

  • IRAS ID

    255165

  • Contact name

    Ewen M Harrison

  • Contact email

    ewen.harrison@ed.ac.uk

  • Sponsor organisation

    The University of Edinburgh

  • Clinicaltrials.gov Identifier

    AC18132, Sponsor Reference Number (ACCORD - University of Edinburgh); CRD 18140 (R534359), Caldicott Approval

  • Duration of Study in the UK

    1 years, 0 months, 7 days

  • Research summary

    Aim: To validate the use of natural language processing (NLP) and machine learning in the assessment of radiology reports.

    Radiology and pathology reports are structured documents which describe findings of investigations such as ultrasound scans. Many reports will document diagnoses, for example, gallstones. The reports can therefore be used to identify patients who have a certain disease either for clinical or research purposes. In this study, the reports can be used to identify patients with gallstones and then to assess whether they develop other health problems such as cardiometabolic disease.

    NLP can automatically determine whether certain features are present in text. An NLP algorithm can assess reports and decide whether a particular diagnosis such as gallstones is present.
    NLP can be used to rapidly assess thousands of reports in a few minutes and determine the number of reports which state that gallstones (or any other conditions) are present. Machine learning offers similar abilities but with the possibility for improvements in accuracy of the algorithm.

    We plan to use NLP to assess ultrasound and MRI scan reports for the presence of gallstones. We then plan to compare the classification of each report with a human classification and machine classification of the report to determine the accuracy of the techniques.

    We then plan to create a dataset of patients with and without gallstones which can be evaluated for their association with other diseases such as cardiometabolic disease.

    Finally we plan to use the techniques on other diseases such as gallbladder cancer and gallbladder polyps to ensure that the algorithms can be used for more than one disease.

  • REC name

    South East Scotland REC 01

  • REC reference

    21/SS/0003

  • Date of REC Opinion

    18 Jan 2021

  • REC opinion

    Favourable Opinion