A Machine Learning Model to Identify and Classify Fractures From X-ray

  • Research type

    Research Study

  • Full title

    Developing AXIS, an Automated X-ray Interpretation System which will detect and grade fractures from plain radiographs (x-rays) using new machine learning algorithms.

  • IRAS ID

    294586

  • Contact name

    Philip Sauve

  • Contact email

    philip.sauve@porthosp.nhs.uk

  • Sponsor organisation

    University of Portsmouth

  • Clinicaltrials.gov Identifier

    TECH 41626, TECH 41626

  • Duration of Study in the UK

    0 years, 5 months, 31 days

  • Research summary

    We are developing our software application in collaboration with the department of Future & Emerging Technologies, University of Portsmouth (UoP).

    Our application will use two AI algorithms to automatically detect and classify fractures from plain radiographs (x-rays).

    We will use approximately 2000 anonymised plain radiograph DICOM images per body part studied, distal radius fractures initially, exported from the picture archiving and communication system (PACS) at PHU as JPEG images. These images will be used to train a transfer learning algorithm. These newer algorithms have recently been shown to have better results than convolutional neural network (CNN) algorithms in some applications. The algorithm will be trained to detect the distal radius on the anteroposterior and lateral images as the region of interest (ROI). Once trained the images are deleted. This type of algorithm has been proven to be very good at image classification (fracture detection). The algorithms will then be tested on a further 500 previously unseen images, consisting of images with and without distal radius fractures.

    The diagnostic performance of the algorithm will be assessed and the accuracy, sensitivity, specificity and Younden Index compared to standards in the medical literature.

    We will then combine this algorithm with an additional knowledge-based system (KBS) algorithm. This KBS will be trained using human clinical expertise (consultant orthopaedic surgeon) based on the fracture type.

    This will produce an automated system for interpreting plain radiograph images (x-rays) and recommending treatment.

  • REC name

    North West - Greater Manchester Central Research Ethics Committee

  • REC reference

    21/NW/0084

  • Date of REC Opinion

    19 Mar 2021

  • REC opinion

    Favourable Opinion