Multi modal image-to-image translation for drug discovery

The GlaxoSmithKline Centre for Optical Molecular Imaging (COMI) led by Professor Boppart at UIUC enables GSK to gain access to cutting edge in-vivo and in-vitro imaging modalities 5-10 years before they are commercially available. The techniques developed at COMI are centred on label-free imaging, which means that no chemical labelling is needed. This both reduces the potential for labels interfering with the processes that we are trying to observe and enable live in vivo imaging of biological processes in real time. This capability is central to addressing the three pillars, enabling longitudinal studies over the time-span of a disease's progression in a single sample. COMI's unique instrumentation further allows multiple imaging techniques to be deployed in rapid succession, giving a multi-modal view in which each modality reveals different information about the sample.
Multi-modal real-time imaging across multiple samples and multiple time-points generates enormous quantities of data. It is not possible for humans to manually screen this, and automated image analysis techniques are necessary. Dr Styles at University of Birmingham (UoB) is a leading expert in this area, and this proposal seeks to initiate a collaboration between Prof Styles and Prof Boppart, with extensive support provided by GSK.
The first part of this project will be to produce "summary" images of a dataset that can be visually interpreted by a human expert. To do this, we will use recent advances in image-to-image translation in which a transformation between two image types is learned in both directions (ie predict type A from B, and predict B from A). We will initially focus on generating synthetic histopathology image (H&E stained) which are very familiar and easily interpretable by human experts (pathologists). Later, we will extend this to learn transformations between modalities. This will then allow us to compare a real image of type B with a synthetic image of type B generated from the corresponding image of type A. This can then be used to identify differences between the synthetic and real images which can be identified automatically through image comparison techniques without any need for human intervention.
In addition to enabling a new collaboration between Dr Styles and Prof Boppart, this project will also start a new collaboration between UoB and GSK with potential for further future engagement and collaboration. The project links well to UoBs strengths in both image analysis/computer vision, and in computational life science that together were the foundation of our membership of the Alan Turing Institute. This project links both of these areas, and this project could lead to further engagement with the Turing. For example, a Turing data study group in the area of this proposal has already been discussed with GSK. The project also links well to the future plans of UoB's School of Computer Science which is seeking to expand its activities in artificial intelligence and in computational life sciences.
Dr Styles will benefit from being able to work on some unique state of the art data that is only available from Prof Boppart's lab. This will enable him to develop new approaches to image understanding that will form the basis of high impact joint publications, and the involvement of GSK means that there is long-term potential for the outcomes of the work to generate impact which could form the basis for a future impact case study.
Prof Boppart will benefit from access to both the image analysis expertise in Dr Styles' group, and also to the high performance computing infrastructure at UoB of which this project will make extensive use. The insights gained from advanced analysis may also serve to guide future development of the imaging technology in order to improve their ability to identify key features of the data that are found to be important biomarkers that reflect disease state and drug effect.