[home] [Personal Program] [Help]

---

Session: Cardiac Diagnostics
Session starts: Friday 25 January, 10:30
Presentation starts: 11:45
Room: Lecture room 559

---

Vikas Gupta (Leiden University Medical Center/Delft University of Technology)
Martijn van de Giessen (Leiden University Medical Center/Delft University of Technology)
Hortense Kirisli (Erasmus MC, Rotterdam)
Sharon Kirschbaum (Erasmus MC, Rotterdam)
Emile Hendriks (Delft University of Technology)
Wiro Niessen (Erasmus MC, Rotterdam)
Boudewijn Lelieveldt (Leiden University Medical Center/Delft University of Technology)

Abstract:
To detect perfusion abnormalities at an early stage of CAD, myocardial perfusion is often assessed by analysing cardiac MR perfusion (CMRP) images. A combination of rest and stress-induced perfusion allows assessing the ability of the heart to adapt to physical exercise, quantified as the myocardial perfusion reserve index (MPRI). However, especially in stress MR acquisitions, the inability of a patient to breath-hold may lead to misalignments between subsequently acquired frames and MPRI, which is based on dynamic contrast uptake (upslope), cannot be measured reliably. Here, we propose a novel motion correction method which is especially aimed at robustness. Motion artifacts manifest themselves as sudden intensity changes over time and show up as high frequency content. We propose to minimize this high frequency content directly by translating all the frames in the sequence, thereby removing the motion artifacts. By writing the necessary discrete time Fourier transforms as matrix multiplications, an efficient solution strategy is derived. A dataset comprising rest and stress images (MRI, 1.5 Tesla) from 10 patients with suspected CAD was used to validate the proposed motion correction method. The registration accuracy of the method was assessed based on annotated myocardium contour locations and clinically relevant parameters (relative upslope, MPRI). The proposed method is also compared to an existing method based on independent component analysis (ICA) [1]. Mean displacements in the non-registered sequences were 2.46 (rest) and 4.85 (stress) pixels (average pixel size: 1.52 mm isotropic). For the proposed method (FT), these decreased to 0.15 and 0.23 pixels, respectively. However, for the ICA based method these were about 1.76 and 5.08 pixels, a motion increase for the stress sequences. Rest and stress upslope parameters of the proposed method (FT) and the ICA method were compared to expert annotations and showed good agreement between FT and expert (not statistically significantly different, level P<0.05), while ICA and experts tended to agree less (P=0.026). ICA mainly failed on stress sequences with large motions. MPRI values showed good agreement between FT and experts. With minimal user intervention (ROI selection in 1 frame), sequences of 50 frames can now be registered automatically in 20 seconds compared to approximately 1 minute required by ICA and 10 minutes required for manual annotation. To our knowledge, the minimal user effort combined with the robustness of the proposed method make it feasible for the first time to process stress sequences in a clinical setting and use MPRI in patient care. REFERENCE: Milles, J., van der Geest, R.J., et al.: Fully automated motion correction in first-pass myocardial perfusion MR image sequences. IEEE TMI 27(11) (2008) 1611–21