3D tissue reconstruction and feature identification under DIC microscopy

Understanding how tissues within the body function, in health and disease, is a key aim of biological research. Clearly the microscopic structure of a tissue is vital to its function. For example, how nerve cells are connected together and arranged underlies brain function. How muscles cells are aligned determines how a muscle will contract.

Looking at a section of tissue under the microscope reveals information about its structure but it is limited by the scattering of visible light and thus only the surface can be examined in any detail. This can be overcome, to a limited degree, by using infrared light combined with a specific form of microscopy termed DIC that uses polarised light. This allows the experimentalist to see deeper into the tissue, to a depth of around a tenth of a millimetre, and produces an image that appears falsely three-dimensional because opposite edges of components in the tissue are highlighted or darkened, as if illuminated from the side and casting a shadow. Any deeper structures are not visible as the light is too badly scattered, producing a diffuse image.

DIC microscopy is a standard technique that is routinely used to identify nerve and muscle cells by eye for electrical recording in physiology laboratories. However there is a vast amount of information contained within the images produced by DIC microscopy that cannot be readily used due to the difficulty in training computers to recognize and automatically analyse the false three-dimensional images. Gathering and analysing this information would significantly increase experimental efficiency and thus minimise the quantity of human and animal tissue required for these sorts of experimental procedures.

We propose to optimise the DIC-imaging method using simple and inexpensive modifications of the imaging hardware and then use mathematical and computer analyses to extract information from multiple images allowing for the three-dimensional reconstruction of tissue structure. By utilising the rapid speed of graphical processors designed for game-oriented personal computers, we aim to develop software that will transform the DIC imaging data on-line so that it can be used by the experimenter to efficiently target specific cell types for recording. Such methods will be of great value to academics but also to the medical profession and to industry.

During many experimental tasks the scientifically useful information extracted represents a fraction of that available. Live implementation of advanced data-analyses methods and improved post-experimental analyses of data can significantly increase efficiency, saving valuable resources. However, the implementation of quantitative methods depends crucially on the level of interdisciplinary collaboration in a field. One experimental mainstay in which the data is obviously under-utilised is the imaging of tissue under Differential Interference Contrast (DIC) microscopy.

DIC microscopy is the standard method for identifying cell types for electrophysiological recording in diverse tissues. However, the DIC video feed is rarely preserved, wasting significant structural information. Moreover, there are no live tools as part of the imaging software that could enhance the experimentalist's visual search for particular cell classes or tissue structures. This lack of live or post-experiment image analyses represents a rich target for interdisciplinary collaboration.

This pilot project aims to develop software that saves the ongoing 2D video feed from DIC microscopy, transforms each image into an intensity map amenable to a host of powerful analysis methods and build 3D reconstructions of tissue. Novel aspects of the project are: the development of open-source software linking image acquisition to motorised-stage movement; the use of powerful GPU processors for image manipulation; and the trial of crowd-sourcing through an image-analysis challenge designed to engage the computer-vision community.

The PI and co-PIs are leaders in the fields of mathematical analysis, quantitative image analysis and electrophysiology. The named RAs have published on image analysis and have experience in microscopy, image analysis, software development and hardware modification. We thus represent a strong and capable team well suited to tackle this interdisciplinary project.

The impacts of this research come from: the development and refinement of instrumentation that has the potential to vastly improve the quality of images produced by DIC microscopy; the development of algorithms to extract data from DIC images and produce a full reconstruction of the tissue; the designing of software running on GPUs to produce live 3D reconstructions, thereby increasing experimental productivity; and the provision of the software written and the data generated as open-source scientific resources.

Principal beneficiaries are academics, the medical community and industry. The first two are treated in the academic beneficiaries section, with the long-term benefit to the health of the nation arising from an improved quantitative understanding of tissue structure.

Benefits for UK industry are: the development of imaging hardware that can be relayed to microscope manufacturers and incorporated into future designs; the development of software to link microscope focus, LED selection and stage movement allowing for live 3D reconstruction could be supplied together with commercial software for electrophysiology and Ca2+ imaging; and the development of novel software for 3D-image reconstruction from 2D DIC images that could have applications in related disciplines. Warwick has in place specialist business-engagement staff in Warwick Ventures and the Development Office to ensure relevant results are exploited.

Communication
The PI and co-PIs have a demonstrated record of academic communication through publications, invited conferences and seminars. They also organise a number of wider communication activities, such as workshops at the university and hospitals that bring together academics and health professionals. They regularly participate in outreach activities to the general public, including panel discussions, internet video of popularised research, and regular experimental demonstrations at primary and secondary schools. Funding has been requested on this grant to run an end-of project workshop on image analysis with a challenge to the computer-imaging community to design improved image analyses methods

Training
The University of Warwick has a dedicated staff training section with Warwick Systems Biology Doctoral Training Centre additionally providing an excellent transferable skills programme. The RAs will be encouraged to follow any relevant courses, particularly on Science Communication and team building. General transferable skills of the project include computer and web-based skills, team-working, presentation and high-level writing skills. The specific skills to be learned during the project are also of wide use in the bio-sector and pharmaceutical industry (cellular imaging) and any quantitative sector such as high-finance (mathematical analysis).