Pulse sequence Real-time MRI

Transcription

1 6/7/213 Making exuses Presenting on behalf of Gustavo Mendonça Segmentation of aorti flow in realtime magneti resonane images Gustavo M. Q. Mendonça Joao L. A. Carvalho Undergraduate student, Sientifi Initiation Program Could not be here Forgot adapter plug! Slide preparation time notebook battery life MRI measurement of SVV Motivation Autonomi ontrol over ardiovasular system: Heart reat variability HRV) Blood pressure variability Venous return variability Stroke volume variability SVV) No gold standard for SVV measurement MRI has potential for non-invasive SVV measurement Carvalho et al., ISMRM 27 & 28) Carvalho et al., ISMRM 27 & 28 Slie presription: asending aorta Anterior to main bifurations Phase ontrast MRI Measures blood veloity through imaging plane Requires preise segmentation of aorti flow Pulse sequene Real-time MRI slie seletion Pros: High temporal resolution: 56 ms New image every 14 ms view sharing) Gatehouse et al. MRM 31:54, 1994 Pike et al. MRM 32:476, 1994 RF veloity enoding Gz spatial enoding Cons: spoiling and refousing Gx Low spatial resolution: 3 3 mm² Low image ontrast Gy Diffiult segmentacon 7 ms 1

2 6/7/213 Data aquisition: real-time spiral phase-ontrast MRI morfologia veloidade fluxo a ores região Eah morphologial maps is assoiated to a veloity map -Data aquiredat3t -Spatialres.: 3 3 mm 2 -Temporal res.: 56 ms - View-sharing: new imageevery14 ms - Vel.range: ±2 m/s MRI measurement of SVV Carvalho et al., ISMRM 27 & 28 Valsalva maneuver Limitations morphology veloity olor flow segmentation Previous implementation: Inaurate segmentation Does not adjust to aorti motion/pulsation This work: improves segmentation Proposed algorithm Parameter initialization One time only; graphial interfae 1. Finding the enter of the aorta Before segmentation of eah frame Fully automati, iterative 2. Imagesegmentation Needstoknowwheretheenter oftheaorta is Initialization: restriting searh region Large enough to ontain aorta during entire aquisiton Calulated from manually-presribed radius 1st frame only) Segmentation: model image onstrution hi-pass image R x, y ) 3.2R 3.2R Gaussian model image 2

3 6/7/213 Proposed segmentation algorithm Flow alulation Flow= veloity area -1 binarymask veloity map pixel area Measured flow: normal volunteers Important parameters Important parameters for thresholding the model image Offset level: emphasizes irular shape Adjusting the threshold level Adujsted on the fly Automati; based on flow Overestimation OK Must separate from neighboring flows Threshold level: sets the radius Based on initialization; adjusted frame-by-frame

4 6/7/213 Another very important parameter Finding the enter of the aorta x n, yn ) = x ' ' xn, yn ) = x n+1 n+1 x, y) = x n+1 n+1 s < ε s Crrn Crrn 1 Crr n s Findingtheenter oftheaorta Finding the enter of the aorta: iterative proess Template I and Template II: Shifted to urrent enter Multiplied with low- and hi-pass images Baryenter is measured new enter Proess ends when: shift 5% of radius Representative results Results: good quality images 4

5 6/7/213 Results: medium quality images Results: low quality images Disussion Results: stroke volume Flow is lower near vessel wall inreased venous return 8 stroke volume ml) 7 6 If enter is inorret by 1% of the radius: 5 rest valsalva handgrip breath-holding shorter diastoles flow inorret by 4% redued venous return 5 6 time s) If threshold is inorret by 1% of the radius: ditto better to overstimate than to underestimate) Conlusions Proposed algorithm Model-based approah: segments a Gaussian-like image Attempts to separate the flow, not to segment the lumen Traking and segmentation seems visually good Considerably improved over our first attempts Next: omparison with segmentation by speialists Demonstrated beat-by-beat stroke volume measurement Real-time MRI has strong potential for non-invasive measurement of SVV Obrigado Thank you ありがとう ありがとう joaoluiz@pgea.unb.br 5