DAMLAR

  • Research type

    Research Study

  • Full title

    Development and Assessment of Machine Learning Algorithms in Radiology

  • IRAS ID

    283632

  • Contact name

    Dermot Mallon

  • Contact email

    dermot.mallon@nhs.net

  • Sponsor organisation

    Imperial College Healthcare NHS Trust

  • Duration of Study in the UK

    4 years, 0 months, 1 days

  • Research summary

    Machine learning is a broad set of techniques that enable computers to perform complex tasks. Due to the potential for rapid and objective analysis of images, there has been a massive expansion in the development of machine learning algorithms in radiology.
    Deep learning, specifically convolutional neural networks (CNN), have been particularly successful within radiology. CNNs have been applied in the diagnosis and prognosis of multiple diseases such as multiple sclerosis, haematoma, ischaemic stroke, and brain tumours.
    One of the major drawbacks of deep learning is their propensity to ‘overfit’, which causes the algorithm to learn irrelevant features of the training dataset. Overfitting results in poor performance in external datasets, which significantly diminishes their clinical value.
    With the proliferation of machine learning algorithms for the analysis of radiology imaging from both academic and commercial enterprises, it is important for these algorithms to be rigorously assessed on independent datasets. Algorithms that have been developed to perform the same task should be compared with each other, and, where appropriate, with expert human readers.
    The overarching aim of this study is to compare the performance of different machine learning algorithms used for the analysis of radiological images.

  • REC name

    N/A

  • REC reference

    N/A