| Περίληψη: | Ultrasound super resolution imaging is gaining more and more ground in clinical applications as it provides information of tissue perfusion and knowledge of the haemodynamic properties of a certain region of interest. It is based upon the injection of micro-bubbles capable of reaching small capillaries into the blood stream and their tracking in the temporal domain. The bubbles are non-linear scatterers of the incident sound waves increasing the echogenicity and thus the contrast between blood and local tissue. They are composed by a gaseous inner layer and an outer layer that can be fabricated by a variety of material like phospholipids. Replacing the inner gas with a drug substance, treatment in cellular level is feasible.
The current MB tracking algorithms are divided into deterministic and probabilistic approaches. The first group is primarily based on a nearest neighbour scheme associating bubbles that are the least further apart. Although this approach can yield above average performance, it requires a high frame rate which decreases the penetration depth. The second tracking approach, incorporates appropriate filters to predict the next state of a MB. The majority of them utilize Kalman filter along with a motion model to foresee the future position, velocity and acceleration of the current bubble and link with a bubble that satisfy this prediction.
This thesis considers the parameters used in a previously developed tracking algorithm, which are developed upon hypotheses of the blood flow and attempts to improve the overall performance. Firstly, the pre-defined thresholds are altered based on conclusions obtained by manually collected tracks, which is considered as the most accurate method of linking. After determining a new set of thresholds, a new approach of transforming the parameters into cost functions and then accumulating them into a total cost, is introduced. The performance of the new approaches is assessed via appropriate feature maps and statistical analysis of the produced tracks.
|
|---|
