Quantitative Histopathological Assessment of Fatty Liver Disease

Principal Supervisors: Prof. Jens Rittscher (Institute of Biomedical Engineering, Oxford) and Prof. Rob Goldin (Dept of Medicine, Imperial College London, to be confirmed)

Co-Supervisor: Dr Matt Kelly (Perspectum Diagnostics)

Background

Liver disease is becoming ever more important and deaths from liver disease are rising across the world. While deaths in the UK from cardiovascular disease, lung disease, cancer, diabetes and are all falling, liver disease deaths have increased by 12% from 2005 to 2008. Non-alcoholic fatty liver disease (NAFLD) is a significant contributing factor. While the causes of the disease are preventable obvious signs and symptoms are often only detected as a very late stage. Multiparametric quantitative MRI, which was invented in Oxford and has been commercialised as LiverMultiScan by Perspectum Diagnostics, is one promising technology that will enable efficient early detection. To date, histopathological assessment of liver biopsies is the standard reference for diagnosis. The goal of this research project is to improve our understanding on how histopathological changes correlate with MR based measurements and to lay the foundation for integrating radiology and pathology to improve patient management.

Given the low diagnostic concordance among pathologists interpreting liver biopsy specimens, we proposed to build on the latest developments in machine learning, computer vision and medical imaging to establish quantitative assessment of the biopsy specimens. This approach will not only assist pathologists in making a more objective assessment, it will also enable a more systematic analysis how certain conditions such as fibrosis, inflammation, steatosis, and ballooning relate to the corrected T1 signal. In order to achieve this goal, the project will address the following research objectives:

Establishing a normalcy model. It is well known that sample preparation and variation in staining affect histopathological patterns. We proposed to generate a comprehensive normalcy model that will form the basis for detecting disease related changes.

Quantitative assessment of disease related changes. Existing categorical grading systems are inherently limited. Summarising multiple histopathological changes that can manifest themselves on different anatomical scale in a single grade results in a poor characterisation of the disease state. We will utilise image-based automated analysis methods to develop a more fine-grained and differentiated representation. This method developed will be assisted by a close collaboration with consulting pathologists.

Systematic correlation with MR quantification and special stains. The developed quantitative representation will be used to analyse how factors such as fibrosis, inflammation, steatosis, and ballooning correlate with other phenotypic measures (eg MRI for fat, iron and fibroinflammatory disease quantification, Perl’s staining for iron).

Perspectum Diagnostics will support this study by providing 300 MRIs (with LiverMultiScan quantification) and needle biopsies. In addition, a selected set of live tissue resections will be provided. An anonymized subset will be made available for algorithm development. Prof Rob Goldin will assist the programme by providing clinical guidance and expert annotations.