This site uses cookies. By continuing to browse the site you are agreeing to our use of cookies.

Deep learning for HRCT classification of fibrotic lung disease

  • Research type

    Research Study

  • Full title

    Deep learning for HRCT classification of fibrotic lung disease

  • IRAS ID

    234527

  • Contact name

    Simon Walsh

  • Contact email

    slfwalsh@nhs.net

  • Sponsor organisation

    Royal Brompton and Harefield Foundation Trust

  • Duration of Study in the UK

    0 years, 5 months, 25 days

  • Research summary

    The overall aim of this study is to use a number of different convoluted neural network architectures to provide a computer-based diagnosis in fibrotic lung disease. The study will involve 2 steps.

    1. Training the computer algorithm: First, a number of different convoluted neural network (CNN) architectures will be trained on a training set of HRCT images in JPEG format. The training set will comprise of retrospective, clinically indicated HRCT scans of a patients with fibrotic lung disease performed since 2007 at the Royal Brompton and Harefield Foundation Trust (RB&HFT). These will be collected from the hospital PACS system and stored on an encrypted hard drive. Training of the computer algorithms will take place on a specialised "deep learning" computer owned by the chief investigator (Dr Simon Walsh). What makes this computer specialised is that it uses multiple high specification graphics processor unit cards (GPUs) which are capable of the massive numerical computation required for deep learning. This process is intractable on a standard (even high specification) desktop computer.

    2. Testing the computer algorithm: These CNN's will then be tested classifying the anonymised HRCTs from a separate cohort test cases based on the ATS/ERS/JRS/ALAT HRCT categories for a UIP pattern (usual interstitial pneumonia (UIP), possible UIP and inconsistent with UIP), again, drawn from the PACs of the RB&HFT. Lastly, the best performing algorithm will be tested on a cohort of 150 cases of fibrotic lung disease already scored by 116 radiologists in a previously published (1).

    (1) Interobserver agreement for the ATS/ERS/JRS/ALAT criteria for a UIP pattern on CT.
    Walsh SL, Calandriello L, Sverzellati N, Wells AU, Hansell DM; UIP Observer Consort.Thorax. 2016 Jan;71(1):45-51.

  • REC name

    London - West London & GTAC Research Ethics Committee

  • REC reference

    18/LO/1392

  • Date of REC Opinion

    2 Aug 2018

  • REC opinion

    Favourable Opinion