University of Cambridge > > Cambridge Mathematics Placements Seminars > Optimising the definition of MR-based lung imaging biomarkers

Optimising the definition of MR-based lung imaging biomarkers

Add to your list(s) Download to your calendar using vCal

If you have a question about this talk, please contact Dr Vivien Gruar.


Magnetic Resonance Imaging (MRI) is of interest in pharmaceutical development for its translational potential from animal research into the clinical setting. MRI of the lung is a relatively novel approach for the assessment of the amount of cellularity and oedema (fluid) as indicators/biomarkers of pulmonary toxicity and/or efficacy. The quantification of these lesions is typically assessed by delineating regions-of-interest (ROIs) on the images and applying several filtering techniques to measure the final levels of these biomarkers. The selection of these ROIs (a process known as segmentation) can be achieved automatically using various computational approaches. However, the great variation in noise and image quality in MR data results in automated segmentation often being challenging. In fact, the greatest difficulty with automation is the diseased lung which appears closer in aspect to adjacent tissue (muscle/liver/heart) than to healthy lung. Therefore, manual segmentation remains the principal mechanism for analysis.


We have at our disposal a large historic animal MR dataset (>1000 tomographic images) on more than seven models of disease/toxicity of the lung, from various strains of rats and mice. Furthermore, although the images were acquired on a single preclinical MR scanner, these were made using various multiparametric MR sequences that result in different contrasts and signal-to-noise ratios. The values of six combinations of segmentations and filtering methods will be provided for testing. These methods include: manual, intensity thresholding, and multi-atlas segmentation; combined with: high-pass filtering for blood vessel and noise elimination, and holistic quantification. Animal species, strains, disease model, and MR parameters are available for use as covariates in the analysis.

Aim & Impact of the Project

In this project, we aim to understand the following questions:
  • to what extent can automated segmentation be used?
  • which methods are best suited for which covariates?
  • when is manual segmentation unavoidable?
  • what are the gaps in analysis that will require new segmentation methods?

The understanding of these statements is crucial to improve the efficacy of MRI investigations, introduce common approaches that can be applied across multiple centres, and establish the precise criteria needed to define respiratory conditions such as COPD (Chronic Obstructive Pulmonary Disease), asthma, IPF (Idiopathic Pulmonary Fibrosis) or drug-induced lung injury (DILI) via MRI . A clearer understanding of the use of MRI in this context will help support improvements to the lives of millions of patients affected by respiratory disease.

The student

You will have a clear understanding of machine learning techniques and cloud point analysis of complex datasets. You should have excellent theoretical understanding of variational optimisation and Bayesian statistics, as well as hands-on experience with unsupervised machine learning techniques. Interest in pharmaceutical development, human and animal physiology, medical imaging, and/or computer vision is beneficial.

If you join us, you will be directly contributing to impact the use of automated methods to extract decision-making biomarkers from image supporting medicinal development. Further, you will be able to liaise with a diverse range of scientists from in-vivo biologists to data scientists, and be exposed to cutting-edge science and equipment.”

This talk is part of the Cambridge Mathematics Placements Seminars series.

Tell a friend about this talk:

This talk is included in these lists:

Note that ex-directory lists are not shown.


© 2006-2018, University of Cambridge. Contact Us | Help and Documentation | Privacy and Publicity