Fotorealistisk visualisering av hjertet Fremtidens 3D ultralydavbildning

Ultrasound is today the most frequently used imaging modality for imaging the heart. Ultrasound is cost efficient, the imaging is done in real time and here is no harmful radiation. The main challenge with ultrasound imaging is that the image quality varies significantly between patients. The image quality is also affected by the skills of the operator. The main purpose of the program "Photorealistic visualization of the heart" has been to develop technology and methods as a basis for a completely new generation of diagnostic 3D ultrasound scanners where the imaging is much less depending on patient and operator skills. In todays 3D ultrasound systems, the images are reconstructed in hardware with little flexibility. A key part of the program has been to develop a new system architecture where the 3D image reconstruction is done in software with much more flexibility. There are several challenges related to processing large amounts of data in real time. During the program we have developed a new hardware architecture consisting of new acquisition hardware connected to a high performance industrial PC with a multi core CPU and two high performance graphics boards. The software part of the architecture is a flexible and reconfigurable chain of software modules. One of the software modules contain two new methods to combine information from many overlapping transmits to reconstruct each pixel in the image. One method utilizes this information to retrospectively focus each pixel, avoiding the need for a focus control. The other method analyzes the channel data from the probe to determine if a pixel is likely to be from a real structure or not and adjusts the intensity accordingly. The new architecture has been integrated into a recently launched ultrasound system. The new acquisition methods have been tested by leading physicians in Europe and US and they have proven to make difficult to scan patients easier to diagnose. In ultrasound imaging, the ultrasound waves propagate through inhomogeneous tissue with variations in speed of sound. One of the big challenges in this program has been to come up with an additional real time reconstruction method which effectively compensates for variations in speed of sound through tissue. During the program we have achieved good results also in this part of the program With diagnostic ultrasound imaging, there are also problems related to low signal-to-noise ratio, signal dropouts and regional variations in the tissue. A central part of the program has been to develop new methods which automatically reduces noise, fills in dropouts and compensates for regional differences. Several existing methods have been studied and new methods have been developed. A method for 3D filtering has been implemented in the system and is now integrated into the recently launched system. A new even more efficient method based on Non-Local-Means has also been developed. This method operates at multiple scales of the image. A third method which robustly estimates the orientation of structures in the image and then utilizes this information to enhance structures has also been developed. Furthermore we have looked at several methods which exploit temporal information in the images. Several methods which estimate the motion of the tissue has been studied and three new methods have been developed. Related to this, several objective criteria for assessing image quality have been developed. The methods are also implemented for 3D ultrasound images. The final part of the program is related to visualization of the 3D images. Todays visualization methods are challenging to control and often results in images which are difficult to interpret. The challenge heere has been to find robust and effective methods which results in photo realistic images with significant clinical value. We have investigated different methods for combining depth coloring, shadows and light reflections. One such method has been implemented and integrated into the recently launched ultrasound system. This method has been tested by leading physicians in Europe and US and found to give superior image quality and more clinically relevant details compared to previous methods. The method is in particularly efficient in combination with the new image reconstruction methods. Finally we have looked at ways to improve visualization of blood flow in the heart. This work has resulted in more intuitive and informative mthods to visualize blood flow in the heart

Ultralyd er i dag den mest anvendte avbildningsmodaliteten for hjerteundersøkelser. Ultralyd er kostnadseffektiv, avbildningen er i sanntid og har ingen skadelig stråling. Ulempen med ultralydavbildning er at bildekvaliteten varierer sterkt mellom ulike p asienter. Bildekvaliteten påvirkes også av hvor dyktig operatøren er til å bruke systemet. Hovedhensikten med prosjektet er å utvikle teknologi og metoder som danner grunnlaget for en helt ny generasjon diagnostiske 3D ultralydscannere der avbildning av hjertet i mye større grad er pasient- og operatøruavhengig. I dagens 3D ultralydsystemer er rekonstruksjon av ultralydbildene implementert i lite fleksibel hardware. En vesentlig del av prosjektet er å utvikle en ny type systemarkitektur der rekonstruks jon av 3D bildene er mye mer fleksibel. Det er her mange utfordringer knyttet til å prosessere store datamengder i sanntid. I hjerteavbildning sendes ultralydstrålene gjennom ulike typer vev med forskjellige lydhastigheter og vevet er i tillegg i bevege lse. Dette tar ikke dagens ultralydsystemer hensyn til. En av de store utfordringene i prosjektet vil være å komme fram til en sanntids-metode for ultralydrekonstruksjon som effektivt kompenserer for forskjellene i lydhastighet. Ultralydavbildning har ogs å spesielle problemer knyttet til lavt signal-støy forhold, signal dropouts og regionale variasjoner i vevssignalet. En sentral del av prosjektet er derfor å finne fram til nye metoder som automatisk reduserer støy, fyller inn dropouts og kompenserer for de regionale forskjellene. Den siste delen av prosjektet vil ta for seg problemstillinger knyttet til visualisering av 3D ultralydbildene. Dagens visualiseringsteknikker er krevende å stille inn og gir ofte bilder som er vanskelige å tolke. Utfordringene her er å finne robuste og effektive metoder som gir fotorealistiske gjengivelser med høy diagnostisk verdi. Prosjektet har stort potensiale for å skape en revolusjon innen 3D ultralyd.