CompMath Seminar - Jana Brunatova, University of Groningen
|When:||Th 23-11-2023 11:00 - 12:00|
Tite: Denoising of dual-VENC PC-MRI with large high/low VENC ratios
Abstract: Medical imaging techniques play a crucial role in acquiring non-invasive information about physiological processes within the human body. One such technique is 4D phase-contrast magnetic resonance imaging (4D PC-MRI), which enables the measurement of blood flow velocity fields. However, the accuracy of velocity field estimation is often limited due to the presence of noise in the acquired images. In this presentation, we introduce an improved technique tailored for situations involving significant variability of flow velocities in 4D PC-MRI images. Our approach is based on the Optimal Multiple Motion Encoding (OMME) method, which requires a minimum of two measurements using different velocity encoding parameters (vencs). By performing a single measurement using a large venc, phase wraps in the results can be eliminated, but the noise level remains high as it is proportional to the venc. Conversely, selecting a lower venc reduces the noise level but increases the number of wrapped voxels. By carefully selecting appropriate vencs, the OMME method effectively combines both measurements, resulting in phase wrap-free images with low noise levels. However, using a large high/low VENC ratio introduces another type of noise in the resulting image. To address this, we propose two denoising techniques that exploit the temporal characteristics of the artifacts present in noisy images. Using these methods, we were able to obtain better reconstruction of the velocity field both on synthetic data and on in-vivo data in comparison with the state-of-the-art ODV correction method.