Accelerated MR-Based Motion Field Measurement for PET Motion Correction in PET/MR

• Published in: The Journal of nuclear medicine (1978) Vol. 60
• Main Authors: Han, Paul, Djebra, Yanis, Petibon, Yoann, Marin, Thibault, Ouyang, Jinsong, El Fakhri, Georges, Ma, Chao
• Format: Journal Article
• Language: English
• Published: New York Society of Nuclear Medicine 01.05.2019
• Subjects: Basis functions; Breathing; Data acquisition; Datasets; Field of view; Gating; Health care; Heart; Image processing; Image quality; Image reconstruction; Image resolution; Magnetic resonance imaging; Measurement methods; Medical imaging; Methods; Positron emission; Positron emission tomography; Respiration; Scanners; Sliding; Spatial data; Spatial distribution; Subspace methods; Therapeutic applications; Thorax; Tomography; Uniqueness
• ISSN: 0161-5505; 1535-5667

Background: The intrinsic resolution of modern whole-body PET scanners is about 4 mm, but the practically achievable resolution of thoracic or abdominal PET imaging may be worse than 10 mm due to inevitable cardiac and respiratory motion. PET/MR provides a unique opportunity to address this issue by taking advantage of the high-resolution images with excellent soft-tissue contrasts offered by MR. In the past, MR-based motion field measurement methods have been developed and used to improve the resolution of PET in free-breathing patients. However, these methods often require cardiac and/or respiratory gating and long data acquisition times, limiting their use in clinical applications. This work presents a subspace-based method to accelerate the imaging speed of MR-based motion field measurement for PET motion correction in PET/MR. Methods: Our method leverages a unique property of dynamic MR images, known as partial separability [1], to accelerate the imaging speed of MR-based motion field measurement. We represent a dynamic MR dataset as partially separable (PS) functions: The true signal ρ(x,t) is expressed as the sum of Ul(x)Vl(t) as l goes from 1 to L, where Vl(t) are the temporal basis functions capturing the underlying dynamics of the data, Ul(x) are the corresponding spatial coefficients determining the spatial distributions of the data, and L is the model order (a small number in practice). The PS model indicates that the high-dimensional dynamic MR images reside in a low-dimensional subspace, which significantly reduces the number of degrees of freedom of the data. Furthermore, it enables a special data acquisition scheme (as shown in Fig.1A) for sparse sampling, where two datasets are acquired in an interleaved fashion: a low-resolution "training" dataset to determine the temporal bases Vl(t) and a sparsely sampled "imaging" dataset to estimate the spatial bases Ul(x). The image reconstruction is done by fitting the PS model to the measured k-space data. To validate the performance of the proposed method, we performed MR scan on healthy volunteers (approved by our local IRB) on a whole-body MR scanner (Trio, Siemens Healthcare, Erlangen, Germany). The imaging parameters are as follows: field-of-view (FOV)=360×304×120 mm3, resolution=1.875×1.875×5 mm, TR/TE=4/2 ms, total acquisition time = ~5 min. Free-breathing continuous MR acquisition without any gating was performed as shown in Fig.1A. Results: Representative reconstruction results are shown in Fig.1B. Images were reconstructed using the proposed method and the conventional sliding window methods, respectively, with a frame rate of 0.22 s/frame. Our method produced high-quality dynamic 3D MR images without noticeable motion artifacts as compared to the sliding window method. Note that only 44 phase-encoding lines were acquired per frame, which translates to 1.13% of the fully sampled data at the Nyquist rate. Conclusions: The partially separable model can be used to accelerate the imaging speed of MR-based motion field method. The proposed method can improve the practical utility of MR-based motion correction of PET in PET/MR.