scAIphoid

• Research type

  Research Study

• Full title

  Development and validation of a deep learning algorithm to diagnose scaphoid fractures on radiographs.

• IRAS ID

  296655

• Contact name

  Dominic Furniss

• Contact email

  dominic.furniss@ndorms.ox.ac.uk

• Sponsor organisation

  University of Oxford

• Duration of Study in the UK

  5 years, 0 months, 1 days

• Research summary

  Background:
  The scaphoid bone is a commonly broken small bone in the wrist. It has an unusual blood supply, which means that if a scaphoid fracture is not diagnosed and treated promptly, this can lead to life-changing complications such as chronic pain or collapse of the wrist joint. Ultimately, patients may need major salvage surgeries. The diagnosis of scaphoid fractures is difficult, and around 3 in 10 initial x-rays are mistakenly reported as normal. MRI scans are more accurate than x-rays, but many hospitals do not have access to timely MRI to aid diagnosis. Preliminary research suggests that artificial intelligence (AI) algorithms can accurately detect very subtle (occult) fractures on x-rays.

  Research questions:
  1. Can an AI algorithm help clinicians improve their accuracy in diagnosing scaphoid fractures on x-rays?
  2. Can an AI algorithm detect occult scaphoid fractures on x-rays and help to prioritise patients for MRI scans?

  Methods and outcomes:
  We will collect x-rays, CT and MRI scans of the scaphoid bone in OUH. We will enlist the help of expert musculoskeletal radiologists to label the x-rays with the correct final diagnosis. We will use some of these x-rays to train an AI algorithm, and some to test the algorithm after training. We will compare the diagnostic accuracy of the AI on its own, Emergency Department (ED) clinicians on their own, and ED clinicians working with AI.

  Dissemination:
  We will involve patients throughout the research process, using the NDORMS “Meet the Researcher” and “OPEN ARMS” initiatives to help design the research, gather feedback, and explain our results. A close patient partnership will run in tandem to inform development of the research at each step. We will publish our research in scientific journals and present at professional conferences, and if the work is successful intend to distribute the software widely within the NHS.
  

• REC name

  London - Camden & Kings Cross Research Ethics Committee

• REC reference

  22/LO/0049

• Date of REC Opinion

  4 Feb 2022

• REC opinion

  Favourable Opinion