Volume rendering is a well known technique for visualizing volumetric data, and is commonly used in many areas of science. In four dimensional cardiac ultrasound (3D + time), volume rendering is used to present the user with a view of the three dimensional structures of the human heart in real-time. However, ultrasound suffers from imaging artifacts including high gray-level variance and dropouts. This makes it challenging to generate high quality volume renderings.
This thesis has focused on how to auto-optimize volume renderings by locally applying adaptive opacity transfer functions and dynamical anatomical cropping, based on segmentation results. In addition, stereoscopic visualization through newly available technology for the gaming marked has been investigated and evaluated.
Locally applied adaptive opacity transfer functions (LOTF) has for the first time been tested out. The method is based on estimating the opacity function parameters, like the slope position and steepness locally from the volume data. Both a method using per ray edge detection, based on a transition criterion, and a method estimating the per ray mean and standard deviation (statistical thresholding surface) has been proposed and investigated. To regularize the LOTF calculations, methods incorporating segmentation results has also been designed and tested.
Auto-optimized methods for dynamical cropping defined by model based segmentation has been proposed. This gives a new way of cropping ultrasound volume renderings. Two approaches has been investigated, one that applies volume depth peeling (hard segmentation crop), and one that applies a novel local opacity weighting scheme for creation of a soft border between included and excluded data (soft segmentation crop).
Stereoscopic renderings of ultrasound based on liquid shutter glasses technology has been tested out and evaluated. Standard tests for fine depth perception and tailored test for depth encoding has been applied. A stereoscopic setup gives the opportunity for different ways of rendering ultrasound data. The method proposed here applies a high transparency rendering scheme in combination with stereoscopic visualization. The setup also makes depth encoding in stereo available.
Among the proposed auto-optimization methods the LOTF based on the statistical thresholding surface shows the most potential. The method removes low level echoes from the rendering, making it bright and sharp. At the same time it also preserves structures of interest. The method also shows good results when we restrict the estimation to segmentation results. In fact, the expert users reported this method to perform better than the global OTF in some cases.
There has been found no statistical difference between the anaglyph and liquid shutter glasses stereoscopic techniques when it comes to fine depth perception. Also the introduction of depth encoding in stereo showed no statistically significant improvement compared to a grayscale rendering. However, experts on image quality favored the liquid shutter glasses technique with depth encoding included, compared to an anaglyph grayscale rendering. The reason for this, may be the result of other parameters than the tested depth perception. Among these we have pure aesthetics, less ghosting and eye strain.