State estimation in dynamic MRI
| Autor: | Marko Vauhkonen, Joanna K. Huttunen, Ville Kolehmainen, Ville-Veikko Wettenhovi, Olli Gröhn, Mikko I. Kettunen |
|---|---|
| Rok vydání: | 2018 |
| Předmět: |
Computer science
Heuristic (computer science) Noise reduction Recursion (computer science) 02 engineering and technology Kalman filter 021001 nanoscience & nanotechnology 01 natural sciences 010101 applied mathematics Temporal resolution Sliding window protocol Dynamic contrast-enhanced MRI Golden angle 0101 mathematics 0210 nano-technology Algorithm |
| Zdroj: | 2018 IEEE Nuclear Science Symposium and Medical Imaging Conference Proceedings (NSS/MIC). |
| DOI: | 10.1109/nssmic.2018.8824307 |
| Popis: | We propose a state estimation approach, enhanced with prior information, for dynamic magnetic resonance imaging (MRI). In state estimation the time series of system states are estimated, utilizing the observation and state evolution models and the dynamic time-varying measurements. In this work the state estimation algorithm is the Kalman filter (KF), where the state estimation formulation is complemented with an anatomical structural prior. In our previous work we used two different methods to include prior information to KF. In the first method the observation matrix was augmented with anatomical prior information and in the second method a heuristic total variation (TV) denoising was applied to each Kalman recursion. While the TV method was faster and better than the augmented one, the TV method lacked a sound mathematical background. In this work, we include the prior information in the state evolution model allowing for significantly faster computation compared to the previous augmented method, yet retaining the solid mathematical background and the same quality of the estimate.The proposed approach is evaluated using simulated functional MRI (fMRI) data and experimental, golden angle sampled, small animal dynamic contrast-enhanced MRI data from a rat brain. When using a state estimation method, we can improve the temporal resolution compared to conventional methods by using only one spoke of radial data for each estimate. We compare the results to the augmented Kalman filter and a sliding window method. The results show that the proposed method produces equally good results when compared to the augmented Kalman filter, but allows for over five times shorter computational times. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|