For Students
Projects
Subject-Independent Arm Motion Classification For Stroke Rehabilitation
Suppose we want to know, in high level descriptive terms, what a person is doing with their arm, but all we have is low level measured data describing tri-axial acceleration and gyroscopic orientation. Thus we require a way to accurately match sequences in this data stream to these high-level categories. In this project we will collect a set of data which is generated by people moving their arm according to these categories, train a classifier, then see how accurately it classifies people who did not contribute training data. Applications include non-invasive and non-intrusive evaluation of arm utilisation in stroke survivors.
http://robotics.usc.edu/~maja/publications/hum-wmarc.pdf
http://web.media.mit.edu/~intille/papers-files/MunguiaTapiaIntilleETAL07.pdf
Second-Life 3D Editor For Pre-Surgical Visualisation
Supposing that anyone can be trained to find particular anatomical structures in a set of 2D images, it might be worthwile developing tools to assist the manual construction of 3D structures within the experimentation platform capabilities of Second Life. In this project we will develop a set of training examples to allow someone to learn to identify the facial nerve in a set of image slices in Second Life, create the capabilities for people to construct 3D structures, then ask surgeons to compare these structures with the structures they 'see' in image slices. These visualisations would give surgeons the capability to more quickly assess critical anatomical layout inside a patient prior to surgery than they currently can with 2D image manipulation.
http://www.wired.com/gadgets/miscellaneous/news/2007/07/wiimote
3D Volume Construction For Bone Visualisation
Current patient scanning technologies are 2D, but slice-by-slice viewing is time consuming and not necessarily accurate. In this project we will develop a set of tools that automatically generate 3D bone visualisations from MRI data (at least), making use of our 3D voxel library (Erodavox) and multiple Sensable Phantom haptic control libraries (Reachin and H3D). These visualisations would allow hollow bone models to be quickly generated from image sets.
Fast Local Shape Identification For Cochlear Implant Surgery Simulation
Training
In some interactive applications where an object is visualised and manipulated in 3D, there may be particular requirements for local surface shapes that users should produce. In this project we will attempt to develop techniques for shape identification that are independent of the distribution of the control points of objects. We will present a demonstration of our method on several test objects (in Matlab) and implement the algorithm (on a GPU) within a prototype cochlear implant surgery simulation, where one requirement of correct surgical drilling procedure is the maintenance of flat surfaces.
http://www.ict.csiro.au/staff/Chris.Gunn/videos/BoneDrillingShort.mpg
http://www.cs.wustl.edu/MediaAndMachines/publications/papers/smi05tdg.pdf
I.L. Dryden and K.V. Mardia, Statistical Shape Analysis, Wiley 1998