PhD EASTBIO: Enhancing Consistency in Extracting Brain Imaging Phenotypes from Clinical Images: Addressing Acquisition Variability and Generating Normative Data

This 4-year PhD project is part of a competition funded by EASTBIO BBSRC Doctoral Training Partnership (DTP).

In the rapidly evolving field of medical imaging, the integration and analysis of data from diverse acquisition systems present a critical challenge. Variability in imaging parameters and modalities hampers the consistency and reliability of analyses when combining data from different clinical settings and analysing longitudinal trajectories. Current harmonisation techniques are suboptimal in clinical settings. Phenotypes from images acquired with the most common protocol and under the scanners at major hospitals can bias automated pipelines, unfairly impacting patients imaged with disease-specific acquisition protocols and/or on scanners at health centres.

Another source of bias is caused by the population sample. Different levels of prevalence of diseases that reflect in structural brain changes also bias computational brain analyses. Automated phenotype extraction methods are often developed and validated in relatively homogeneous research datasets, whereas routine clinical imaging encompasses much greater variation in anatomy and pathology. The extent to which this heterogeneity affects measurement reliability, reference trajectories and comparability across datasets remains poorly characterised.

This project has the following aims:

1. Develop and validate robust extraction pipelines for technically challenging imaging phenotypes from heterogeneous routine clinical images: fractal cortical complexity, brain health index, white matter damage, brain volumes, white matter hyperintensities, and infarcts.

2. Quantify the influence of scanner, sequence, imaging modality, and acquisition parameters on each phenotype and develop harmonisation/QC strategies that preserve biological age- and sex-related variation.

3. Derive age- and sex-stratified clinical reference distributions for these phenotypes, with particular emphasis on measures for which large-scale reference values do not currently exist, and compare them with those acquired from research populations.

4. Test the reproducibility and clinical informativeness of these reference trajectories across independent datasets and diagnostic groups.

The student will receive training in medical image processing, big data analyses, working in the safe heaven, and advanced statistics. The principal supervisor will give day-to-day supervision in brain image processing and analysis methods, and the harmonisation of brain imaging data.

This opportunity is open to UK and international students and provides funding covering stipend and UK level tuition fees.