Techniques and software architectures for medical visualisation and image processing

This thesis presents a flexible software platform for medical visualisation and image processing, a technique for the segmentation of the shoulder skeleton from CT data and three techniques that make contributions to the field of direct volume rendering. Our primary goal was to investigate the use of visualisation techniques to assist the shoulder replacement process. This motivated the need for a flexible environment within which to test and develop new visualisation and also image processing techniques with a medical focus. The Delft Visualisation and Image processing Development Environment, or DeVIDE, was created to answer this need. DeVIDE is a graphical data-flow application builder that combines visualisation and image processing techniques, supports the rapid creation of new functional components and facilitates a level of interaction with algorithm code and parameters that differentiates it from similar platforms. For visualisation, measurement and pre-operative planning, an accurate segmentation from CT data of the bony structures of the shoulder is required. Due to the complexity of the shoulder joint and the fact that a method was required that could deal with diseased shoulders, existing techniques could not be applied. In this thesis we present a suite of techniques for the segmentation of the skeletal structures from CT data, especially designed to cope with diseased shoulders. Direct volume rendering, or DVR, is a useful visualisation technique that is often applied as part of medical visualisation solutions. A crucial component of an effective DVR visualisation is a suitable transfer function that assigns optical characteristics to the data. Finding a suitable transfer function is a challenging task. We present two highly interactive methods that facilitate this process. We also present a method for interactive direct volume rendering on ubiquitous low-end graphics hardware. This method, called ShellSplatting, is optimised for the rendering of bony structures from CT data and supports the hardware-assisted blending of traditional surface rendering and direct volume rendering. This characteristic is useful in surgical simulation applications. ShellSplatting is based on the object-order splatting of discrete voxels. As such, maintaining a correct back-to-front or front-to-back ordering during rendering is crucial for correct images. All existing real-time perspective projection visibility orderings show artefacts when splatting discrete voxels. We present a new ordering for perspective projection that remedies these artefacts without a noticeable performance penalty.Electrical Engineering, Mathematics and Computer Scienc