Community Research and Development Information Service - CORDIS

Matching Methodology


The "Matching" work in I-Match concentrates on the development of techniques to optimise the selection and matching of an Assistive Technology (AT) device to AT users. This matching of device to user can be supported by measuring both the functional characteristics of the device and capturing the characteristics and the skills of the users.

Correctly matching appropriate Assistive Technology (AT) devices and AT interfaces to users with disabilities is a key aspect involved in their eventual take up and adoption of Assistive Technologies. Another factor in their continuous use relates to the trial-ability of the AT devices, with studies showing a higher adoption and use if the users were able to experiment and train with the devices before adoption.

One of the principle objectives of I-Match has been to use State of the Art measurement and simulation technologies to enable people with functional impairments to choose the optimum interface systems for AT devices. To this end a virtual reality environment was developed so the users can evaluate and test several devices, including a MANUS, and HANDY-1 and a powered wheelchair with various interface controls e.g. Haptic device, Switches, and Joysticks etc.

Also in line with this objective software was developed, for use with a haptic PHANToM® device, that would provide a series of tests and simulations that would allow gauging of upper limb performance. The output from these simulations, tests and training sessions can be used to generate reports that could be used by Assistive Technology Assessors to recommend an appropriate interface device for a user.

This also involves the development of standardised tests and testing procedures for evaluating the user experience with the devices. Comparison of this new method of measuring upper limb performance with traditional measures like Fugl-Meyer, Barthen and the Rivermead motor assessment score are also an aspect to this study.

The final aspect of the project is the development of a decision support system that will aid Assistive Technology Assessors in recommending an appropriate device to the user. This will take the form of two systems, a Case based reasoning system and a rule based system decision support system.

Case-Based Decision Support System

The system uses the libraries, CBML and Fionn, developed by the Machine Learning Group in Trinity College Dublin using the Java programming language; the I-Match CBR system of Assistive Technologies therefore employs Java as the development platform to facilitate integration of these components. The Graphical User Interface, GUI, uses the Java Swing API system.

The I-Match CBR System of Assistive Technologies has four principle components; the Case Base Manager, the Database Manager, the GUI and Feature Selection Component. These systems use the Fionn Framework and the CBML system. The FIONN Framework is designed for the development and testing of Machine Learning algorithms. Included in FIONN is support for all the basic aspects of a CBR system. FIONN works with an XML system called Case Based Mark-up Language, CBML, which represents cases in the system. Fionn is independent of the classifier or evaluation technique used; this allows various classifiers to be changed e.g. from a k-NN classifier to using a Naive Bayes classifier.

The CBML system comprises two aspects. The first CBML format developed in XML that represents cases in a standardised and independent manner. The second aspect of the CBML System is an API developed for manipulation of CBML files. The first aspect of CBML, namely the CBML XML files are independent of application code, so any system that can read XML can use CBML. The CBML API is an implementation to simplify the job of working with CBML files.

The system was evaluated on a device selection task, which involved selecting a joystick for a user from two choices. The evaluation was done on a test set of 26 users with the system recommending a device for each user using a leave-one-out methodology. While the evaluation was successful with approximately 80% accuracy this was achieved with considerable effort expended on collecting good quality training cases. Our conclusion is that simply collecting the 'naturally occurring' cases in a clinic and using these, as training data for a decision support would yield poor quality solutions. Because of this, we do not propose to continue with this decision support application at this stage. Future work at TCD is unlikely as the PI responsible for the work is no longer at the institution. It is possible that the work could be continued by another I-Match partner.

Related information

Reported by

Follow us on: RSS Facebook Twitter YouTube Managed by the EU Publications Office Top