Machine learning and outcome prediction in paediatric brain tumours

  • Research type

    Research Study

  • Full title

    The integration of multi-parametric paediatric brain tumour imaging with biological, clinical and histo-pathological features using machine learning to maximise outcome prediction and biomarker discovery

  • IRAS ID

    257763

  • Contact name

    Stavros Michael Stivaros

  • Contact email

    stavros.stivaros@manchester.ac.uk

  • Sponsor organisation

    University of Manchester

  • Duration of Study in the UK

    2 years, 11 months, 31 days

  • Research summary

    Research Summary

    Brain tumours are the second most common childhood cancer, but the highest cause of cancer related death in children. They create a challenging patient group due to the large number of different tumour types and volume of information associated with each patient. Statistically, it is difficult to predict recurrence and survival rates on an individual patient basis.

    Advanced brain MRI techniques generate detailed images, describing both the anatomical location and physiological behaviour of these tumours. Currently, we do not integrate these detailed findings with clinical information including tumour pathology and patient outcomes. Research requires a new approach to tackle this data, allowing it be processed in a useful manner. Machine learning, or artificial intelligence, techniques may be the solution to this problem.

    We are aiming to establish whether linking MRI findings with corresponding biological and clinical information can identify recurrence and relapse patterns for individual tumours. This will provide patients with personalised information when commencing treatment and also allow rationalisation of surveillance MRI scans. This will ensure that potential tumour relapses are identified and treated appropriately, the risks of excess scans are avoided and the financial burden of unnecessary scans is reduced.

    Imaging, biological, surgical and clinical data will be collected from patients who have previously been diagnosed with a brain tumour and undergone treatment. This is a retrospective study; patients will not be subjected to any additional treatments or investigations.

    We will trial machine learning techniques, assessing whether recurrence, relapse and survival patterns can be identified from our data. Successful techniques will be tested against new data sets to assess whether results are reliable and reproducible.
    Eligible patients will be recruited from Royal Manchester Children’s Hospital, Leeds General Infirmary and The Royal Victoria Infirmary Newcastle.

    This study is funded by Children with Cancer UK.

    Summary of Results

    My study has focussed on improving how we predict outcomes in paediatric brain tumour (PBT) patients, particularly focusing on how we incorporate MRI, pathology and clinical into our prediction modelling. PBTs are a notoriously challenging group of patients to statistically predict outcomes for due to the overall relatively low number of cases. To try and tackle this, we have trialled the use of novel machine learning and artificial intelligence modelling techniques to create predictive models.

  • REC name

    London - Surrey Borders Research Ethics Committee

  • REC reference

    19/LO/1975

  • Date of REC Opinion

    2 Mar 2020

  • REC opinion

    Further Information Favourable Opinion