The School of Informatics has been awarded funding from the Wellcome Foundation to undertake a research project entitled "Dermofit: A cognitive prosthesis to aid focal skin lesion diagnosis". The principal investigators are Prof. Robert Fisher and Prof. Jonathan Rees. The project funding is until July 31, 2011.

Context

The research proposed here will take place in the Machine Vision Unit of the Institute of Perception, Action and Behaviour (http://www.ipab.informatics.ed.ac.uk). The Institute investigates how to link, in theory and in practice, computational action, perception, representation, transformation and generation processes to external worlds. Research activities include computer vision, mobile and assembly robotics and visualization. The Machine Vision Unit has a long history of research with three-dimensional data, including 3D sensing, surface description, object recognition and automatic model acquisition. More details can be found at http://www.ipab.informatics.ed.ac.uk/mvu/.

The Institute of Perception, Action and Behaviour is a research institute in the School of Informatics at Edinburgh University. The University is internationally known for its School of Informatics, which is now the largest category 5 academic research and teaching department of Informatics in the United Kingdom. At present there are about 75 academic staff, 100 contract research staff and about 380 postgraduate MSc and PhD students. Edinburgh was the only University in the UK awarded the top 5*A rating in Computer Science in the 2001 Research Assessment Exercise (the latest review is still in progress). Edinburgh is also the UK's biggest research group in Informatics. Research activities cover almost every aspect of Informatics. More details about the School and its research can be found here.

Project Background

One goal of the project is develop a tool that will allow non-experts to diagnose skin lesions by taking advantage of the ability of humans to make visual matches even when they are not able to describe the lesions (using words) in a consistent way. In order to attach semantics to images we will have to discover the basis of similarity between different lesions such that we can categorise and index a database in such a way that lesion images that are similar are tagged as similar. If the search space can be structured in this way then it becomes possible for non-experts to search it efficiently, and match an index case with a tagged-reference case. We will acquire images, construct a user interface, and use iterative testing and user interaction coupled with machine vision and machine learning techniques, to order the database. Just as the pattern of hypertext links reflects a webpage's importance, so does the pattern of clicks from one image to another reveal what users consider as similar. The approach is therefore that of computer based image retrieval. A second project goal is to develop the tool in such a way that it can assist non-experts achieve the correct diagnosis over the web or locally at a PC.

Consider the following: A patient presents in primary care. The practitioner is uncertain of the diagnosis. S/he has access to a large library of skin images. The problem is then one of matching the index case with the appropriate diagnosis-tagged reference image in the library. Of course, since the library is large, and ordered randomly, it is unlikely that a match can be achieved in a practical time. Instead, imagine that the library has been structured such that images that are judged to be similar are found together, and the practitioner can interact with the database, using the structure as a map. This structuring of the database is akin to the way a library uses a catalogue to place books on related topics close to each other. This indexing means that the search path length is greatly reduced. Is it possible to imagine an analogous indexing based on images rather than words? Are there computational analogues to the user's similarity assessment that can bootstrap the machine learning process? Although we will make use of automatic feature extraction and analysis, the novelty of our approach in this context is to bootstrap what is considered 'alike' based on how users move through the database. So, just as the pattern of hypertext links provides information about the 'importance' of web sites (based on how prior users have made choices), the pattern of user clicks provides a measure of likeness that is (intentionally) recorded. If in our patient example, the practitioner is presented with a screen shot of 12 images, s/he will click on the one (or more than one) that s/he considers most like the index image. The database then provides another 12 similar images based on the combination of properties of the lesion previously selected and the pattern of other user's usage. The process is repeated until a diagnostic match is chosen. Each click reveals what the practitioner considers alike. Scale this process up, run it with large numbers of iterations (lots of users) and the search space becomes ordered. Our research questions are: (i) how much order is possible, (ii) can we invent algorithms that take advantage of this structure and (iii) how do we combine automatic feature extraction and analysis, and indexing based on user interaction.

We have conducted a number of pilot studies over the last two years using undergraduate medical students and computing MSc project students. These showed that non-experts were capable of matching lesion images with a high of commonality between observers, thus supporting our intuitions that image-based categorisation is possible. Secondly, we have implemented in Java a browser like interface suitable for later web use. Using an unsophisticated Bayesian classifier of segmented images we have shown around a 84% correct initial classification rate based on image properties (note this was on a narrow range of conditions and we would not extrapolate to a broader context based on this pilot data alone).

Project Tasks

Your projective objectives will include:

1. Assistance with clinical collection of many colour and 3D shape images from a wide variety of skin lesions: this will be one of the primary tasks of the clinical assistant, but you will help advise on technical issues.
2. Mounting these images for public use (on the web): the clinical assistant will assist with this, but one of your tasks will be to develop the web page organisation and infrastructure to support public download of the image data.
3. Development of a variety of colour, shape and texture image properties: Central to the project are the properties that computationally encode the similarity between images that are visually and/or diagnostically similar. Some previous properties extraction algorithms have been developed, but many more exist in the literature and others specific to the distinctive character of the investigated lesions can be developed. Properties that are invariant to the illumination brightness and colour are particularly important. An automatic spot extraction algorithm will be developed (some literature exists for malignant melanoma). This component is likely to require a year of effort over the 3 year period.
4. Machine learning algorithms to discover the most usable features, plus to discover the latent patterns of likeness based on user interaction: Given the likelihood of nearly 100 potential features, a 'best' subset of something like 10 features will be used, based on a feature selection process. A second learning-based process will be based on 'data-mining' what the users consider as similar images, where data is obtained from the web interface, with the goal of identifying what cues users use to judge similarity among the lesions. A third learning-based process will apply the visual similarity cues, plus visual lesion properties, plus live interactive feedback to search the image database for potential image-based matches. This process may have a sample image input by the user, or may be based just on feedback. This component is likely to require a year of effort over the 3 year period.
5. Java interface extension for web use and new capabilities: The existing Java interface prototype will be extended, cleaned up, and made usable over the web.
6. Clinical and web-based diagnostic assessment: experiments will be made to assess how well the algorithms perform at diagnosis, how quickly a user can identify suitable images from the database, and how reliably this can be done, etc.
7. Usability assessment: experiments will be made periodically to assess how convenient the Java user interface is to different categories of users (eg. consultant, GP, nurse, pharmacist, lay-person).

Project Environment and Conditions

Data capture will primarily occur in the Department of Dermatology. The project will use several PCs attached to a departmental file server. Altogether, there are 10 workstations available for use by the vision research group (normally consisting of about 5-10 members, including contract research staff, PhD and MSc students). A 1000+ node parallel system compute server is also accessible by the group. Wherever possible we will use MATLAB for algorithm development and JAVA for web-use within a LINUX/UNIX environment. There is some existing software related to this project. There is one other person researching topics in skin in the group (Li).

The Post

One post-doctoral researcher will be hired for this contract.

The researcher hired will be responsible for undertaking research in the areas listed above, as well as cooperating with other research staff and students working on the project. The specific scientific work will depend on the project schedule, the progress of other members of the research group and abilities of the researcher.

Applicants for the post must have a PhD in an appropriate area, such as computer vision or image processing and should have experience with the MATLAB and JAVA programming languages and good mathematical skills. Experience with image database indexing or medical image processing is highly beneficial.

The post is on the Edinburgh UE07 scale (28290-33780 pounds/annum) and is available from August 1, 2008 until July 31, 2011. Placement for the post is according to experience and qualifications.

Further Job Details

Other research duties will include involvement in the preparation and presentation of demonstrations, some support of the research group's publicity WWW pages, a stint at organising the Institute of Perception, Action and Behaviour's weekly seminar, visiting other vision research sites (mainly in the UK and Europe), the preparation and presentation of papers, attendance at conferences and internal seminars. The applicant will be expected to keep abreast of relevant international research in the area and to prepare and present conference and journal papers. The applicants should be able to produce evidence of the ability to conduct and publish original research.

The team member is also expected to contribute to the general project welfare, such as general software maintenance work, documentation, report writing, graphics coding, etc.

School duties may include some teaching or tutorial work, such as presentation of internal demonstrations, second supervision of undergraduate, MSc and PhD students and development and marking of course work.

The researcher may be asked to attend selected MSc course modules to acquire familiarity with topics relevant to their work duties.

The researcher will work mainly in the new Informatics Forum, recently constructed in the central campus area. This will include the vision group. The Department of Dermatology is a 10 minute walk and you will spend time there assisting with data collection and project meetings.

Several recent surveys placed Edinburgh as one of the most desirable urban locations in the United Kingdom.

Normal Department working hours are 9:00 am to 5:00 pm, Monday to Friday inclusive, for a total of 35 hours per week. Holiday entitlement is six weeks per year, plus public holidays, to be taken during University vacation periods. An optional University Superannuation Scheme is available.

Applications should include a curriculum vitae and the names and addresses of two referees.

Further information is available from, and applications are sent to:

The closing date for applications is May 30, 2008.

In your application, please quote reference number 3009159.

The application form can be found at www.jobs.ed.ac.uk, which also allows an online application procedure.

NOTE: This statement does not of itself constitute a contract or conditions of service.

Back to main page.