Artificial Intelligence to detect early hip replacement failure

  • Research type

    Research Study

  • Full title

    Using Machine Learning to detect or predict loosening of the femoral component of an uncemented total hip replacement.

  • IRAS ID

    265498

  • Contact name

    Richard E Field

  • Contact email

    richard.field2@nhs.net

  • Sponsor organisation

    Epsom and St.Helier Hospital University Hospital NHS Trust

  • Clinicaltrials.gov Identifier

    N/A, N/A

  • Duration of Study in the UK

    3 years, 0 months, 1 days

  • Research summary

    At present, patients with loose hip implants may have pain. The diagnosis can be confirmed by changes seen on serial radiographs; a process which can sometime take months or years because changes can be subtle. Quicker, accurate confirmation as to whether an implant is loose would benefit the patient in this situation. Furthermore, a significant number of patients with pain following hip replacement have radiographs that fail to show a loose implant.

    Early detection/ability to predict loosening, would benefit the rapid development of newer/better designed hip replacements. Currently, the use of newly designed hip replacements need to be followed up for 10 years.

    An alternative option is to use a technique called RSA (Radiostereometric analysis), where metal beads, the size of small seeds, are inserted into the bone around a hip replacement at the time of surgery (9). Using special x-rays (taken from two different directions), these markers can be used to accurately assess whether loosening has taken place. Hip replacements that show no loosening at one year using this technique do not need to be followed up for 10 years. The problem is that this technique involves leaving metal particles in a patient, requires one year of follow-up, and can only be done in a research setting.

    We aim to investigate whether Machine Learning (ML)/Artificial Intelligence (AI) can identify loose implants through non-invasive means. If so, can it do this earlier than humans are able to? Can Machine learning predict which implants will go on to become loose?

    To summarise, assessing minor temporal changes in position of a joint replacement using serial radiographs can be difficult. We hope to refine the method of ML to achieve these comparisons between one radiograph and another, taken at a different time point, with a high degree of accuracy and reliability.

  • REC name

    West Midlands - Coventry & Warwickshire Research Ethics Committee

  • REC reference

    19/WM/0263

  • Date of REC Opinion

    9 Aug 2019

  • REC opinion

    Favourable Opinion