Histo-MRI map

  • Research type

    Research Study

  • Full title

    Learning MRI and histology image mappings for cancer diagnosis and prognosis

  • IRAS ID

    251440

  • Contact name

    Shonit Punwani

  • Contact email

    s.punwani@ucl.ac.uk

  • Sponsor organisation

    Research Governance Manager, Joint Research Office (part of the Research Support Centre)

  • Clinicaltrials.gov Identifier

    NCT04792138

  • Duration of Study in the UK

    3 years, 0 months, 0 days

  • Research summary

    This project aims to exploit recent advances in machine learning to address acute problems in cancer management, most directly prostate cancer. We plan to do so by acquiring co-localised MRI and histology data from 50 prospective prostate cancer patients, in order to develop a revolutionary tool for cancer diagnosis and prognosis based solely on MRI, limiting the need for invasive biopsies.

    The current standard approach of making treatment decisions via biopsy and histology has two key limitations: it is i) invasive and ii) subjective/inconsistent. We will develop computational tools supporting new solutions that resolve both issues.

    Specifically, we aim to enable non-invasive MRI to become the primary diagnostic tool, avoiding a large number of unnecessary biopsies, which carry significant risk of life-changing side-effects, and reserving the procedure for marginal cases. We also plan to relate MR signals to quantitative tissue features to enable more reliable treatment decisions.

    We are uniquely positioned to obtain associated MRI and histology images to support a learning and estimation process from MRI to histological features. Such a mapping can provide invaluable new information for clinical decision making via optimally designed MRI protocols.

    The project involves a number of engineering challenges:
    i) overcoming the intrinsic challenge of aligning images obtained via MRI and histology;
    ii) provide easily understandable measures telling end users how certain cancer predictions made by the trained computer programs are;
    iii) find features in huge data sets that are relevant for cancer characterisation;
    iv) develop strategies that minimse the number of MRI images required by the computer programs for fully non-invasive MRI-based cancer diagnosis.

  • REC name

    London - Queen Square Research Ethics Committee

  • REC reference

    19/LO/1803

  • Date of REC Opinion

    23 Jan 2020

  • REC opinion

    Further Information Favourable Opinion