Dr. Maria Susana Avila Garcia 1, Prof Anne E. Trefethen 1, Prof Sir Michael Brady 2 and Dr Fergus Gleeson 3 Lowering the barriers to Cancer Imaging 1. Oxford e-Research Centre, University of Oxford, UK 2. Dept. of Eng. Science, University of Oxford, UK 3. Radiology, Nuffield Dept. of Medicine, Churchill Hospital, University of Oxford, UK

At Oxford Researchers working in image analysis of colorectal and liver cancer images: –Segmentation –Registration –Image quality improvement. Analysis of medical images is difficult since they are: a)Noisy, b)Highly textured, c)Poor contrast relative to their surroundings.

State of the art Graph Cuts Level Sets Active Shape Models Investing 30% of their research time developing methods, algorithms and tools to support their research... Sometimes they abort the development of such methods because:  It is laborious and time consuming  Do not have all the information required.

Segmentation use case Semi automatic Segmentation Images provided by Dr. Niranjan Joshi using a non- parametric mixture model (NPMM) level sets approach Original Image

Validation Manual Segmentation Ground truth – Set of images with the shapes of interest manually segmented by specialists. Tedious Time consuming

Validation Semi-automatic segmented shapes Ground truth shape Superimposed shapes

And then? Results are shown and discussed with clinicians Deploy algorithms for clinical use UPS!... Need time to develop a suitable interface Clinicians reach a frustratingly low limit to what they can do with reasonable effort with no specialist programmer’s support

Aim alleviate the frustration of non IT users who are not able to analyse and process images with reasonable effort maximise the efficiency of an image analysis researcher  A platform independent framework.  A repository of algorithms with no bounds to specific programming languages.  Access to already existing imaging and visualization toolkits with no bounds to specific programming languages.  Access to the most up-to-date authoritative knowledge.  A framework for rapid development and deployment of applications for clinical use.  Improve mechanisms for manual segmentation  Use of Collaborative visual tools (including multi-touch and interactive surfaces) for multi- user and visual data input.

Cloud Computing Framework Security Various levels of information access to provide security and data confidentiality when needed Provenance contributions of each researcher are registered and the use of their methods and experimental data is acknowledged Web Services Metadata Cancer Imaging Cloud Computing Framework Experiment Manage the concept of experiments where links to various objects can lead the researcher to the information required. Collaboration environment Provide discussion forums to enable communication and collaboration among researchers Metadata Efficient access to the most up-to-date, authoritative knowledge that can serve as metadata

Web-based Application & Virtual Research Environment Metadata Web Services Workflows Code Matlab, C++, Java Experimental data Data Logs Images Additional data Reports Publication list Presentations Image processing & Visualization toolkits User interface tools Scientific Workflow Workbench Enriched Desktop Application WS Cancer Imaging Cloud Computing Framework ? My Experiment Carmen Research Information Centre, RIC Taverna Microsoft Workflow Foundation SciRun IRIS Explorer

Collaborative and Interactive Visual Tools Manual segmentation -Ground truth generation- Graphics tablets Pen tablets Wacom Cintiq 12WX Graphics tablet technique was the most preferred; 60% of the subjects chose the mouse as second preference. In terms of segmentation accuracy and time the graphics tablet technique was the best. The pen tablet technique showed slightly better results than for the mouse.

Collaborative and Interactive Visual Tools Microsoft® Surface Computer DiamondTouch (Photo: Business Wire) (Photo: Mitsubishi Electric Research Laboratories)

Collaborative and Interactive Visual Tools Microsoft® Surface Computer DiamondTouch (Photo: Business Wire) (Photo: Mitsubishi Electric Research Laboratories) Engage clinicians with the use of medical image analysis applications developed by MIA researchers. Improve interactions among MIA researchers and clinicians. –Manual segmentation. –Discussing results. –Enhancing the analysis of images and the sharing of concerns and suggestions.

Challenges The adaptation of existing software. Unbound up toolkits from specific languages. –We will explore various solutions such as the MATLAB® Builder™ NE for Microsoft®.NET framework. Control of contributions made by users. The design and development of suitable user interfaces for clinicians. Promote the access to engineering and computer science academics, and to undergraduate students, –to raise interest in challenges to solve computational and software engineering problems. Engage medical and biomedical science academics and students with the use of image processing techniques –Collaborating with students and researchers of different backgrounds. Link to permanent and secure online archive, –a repository for research materials produced by scholars at Oxford University, to ensure access to a permanent and secure online archive.

Acknowledgements This project is funded by the Technical Computing Initiative of Microsoft Corporation. We thank participants involved within the evaluation of graphics tablets for the generation of ground truth. We also thank MIA researchers at Oxford for their valuable comments during the analysis of requirements for this project, especially Vicente Grau and Niranjan Joshi, as well as radiologists working at Churchill and John Radcliffe Hospitals