Convex Optimized Diffusion Encoding (CODE) Gradient Waveforms for Bulk Motion Compensated Cardiac Diffusion Weighted MRI
Eric Aliotta1,2, Holden H Wu1,2, and Daniel B Ennis1,2

1Radiological Sciences, UCLA, Los Angeles, CA, United States, 2Biomedical Physics IDP, UCLA, Los Angeles, CA, United States


Bulk motion compensated diffusion encoding is critical for accurately measuring diffusion in the heart. However, the diffusion encoding gradient waveforms required to suppress bulk motion artifacts can extend TE and limit SNR. We have developed a Convex Optimized Diffusion Encoding (CODE) framework to design time-optimal, motion compensated diffusion encoding gradients that remove sequence dead times and minimize TE. CODE gradients were designed and implemented for cardiac DWI on a 3.0T clinical scanner in healthy volunteers and patients. CODE reduced bulk motion artifacts compared with conventional monopolar encoding.


To implement and evaluate Convex Optimized Diffusion Encoding (CODE) gradient waveforms for bulk motion compensated cardiac Diffusion Weighted MRI (cDWI) with minimum TEs.


cDWI has the potential to characterize cardiac microstructure without the need for a Gadolinium-based contrast agent (GBCA), which is important for the large number of patients with poor renal function requiring evaluation by cardiac MRI[1]. The clinical utility of cDWI, however, has been limited by severe sensitivity to cardiac motion. Recent reports of motion compensated (MOCO) diffusion encoding gradients with nulled first (M1) and second (M2) moments have demonstrated robustness to bulk cardiac motion[2, 3], but they necessarily increase the echo time (TE) compared to monopolar encoding (MONO). Increased TEs reduce SNR and limit spatial resolution. We have developed a MOCO cDWI sequence that employs Convex Optimized Diffusion Encoding (CODE) to reduce bulk motion sensitivity and shorten TE compared to existing MOCO schemes.


Gradient Design: CODE diffusion encoding gradients were designed using convex optimization to determine the M1 and M2 nulled gradient waveform that minimizes TE for a target b-value while conforming to hardware constraints (GMax=74mT/m and SRMax=50T/m/s) and pulse sequence timing.

Healthy Volunteer Imaging: Healthy volunteers (N=10) were scanned on a 3.0T scanner (Siemens Prisma) after providing written informed consent. High resolution cDWI were acquired in the left ventricular (LV) short-axis with b=350s/mm2, 1.5x1.5x5.0mm spatial resolution, 2x GRAPPA acceleration, three orthogonal diffusion encoding directions and three signal averages in a single 15-heartbeat breath hold. Both MONO (TE/TR=67ms/1R-R) and CODE-M1M2 encoding (TE/TR=76ms/1R-R) were acquired at eight subject-specific cardiac phases distributed across systole and diastole.

Reconstruction and Data Analysis: Apparent diffusion coefficient (ADC) maps were reconstructed for each cardiac phase. Motion corrupted voxels were identified by ADC values exceeding 3.0x10-3mm2/s (the diffusivity of free water at 37°C, a thermodynamic upper bound for soft tissues) in the LV. The mean myocardial LV ADC and the percentage of motion corrupted LV voxels were then calculated for each phase. Statistical analyses were performed using t-tests with Holm-Sidak post hoc corrections.

Clinical Imaging: Patients (N=5) undergoing routine clinical cardiac MRI exams were also scanned after providing written informed consent. cDWI were acquired before and after the injection of a GBCA (Gadovist, Bayer Healthcare) using the CODE-M1M2 cDWI protocol at a single early systolic phase (100ms delay from the QRS complex via ECG). Mean myocardial LV ADC was calculated after manual segmentation for each patient.


The TE for CODE-M1M2 (TE=76ms) was 19% shorter than modified bipolar MOCO[3] (TE=94ms) for 1.5x1.5mm in-plane resolution and b=350s/mm2 (Figure 1). With MONO the mean ADC values were significantly corrupted (>3.0x10-3mm2/s, p<0.004) for 50% of the cardiac phases whereas 0% of the cardiac phases were corrupted with CODE-M1M2 (p=N.S.) (Fig. 2B). CODE-M1M2 measured significantly lower mean ADCs than MONO (1.9±0.3x10-3mm2/s vs. 3.8±0.6x10-3mm2/s, p<0.007) and fewer motion corrupted voxels (14±14% vs 67±21%, p<0.0006) in 100% of the cardiac phases (Fig. 2C).

The clinical CODE-M1M2 cDWI scans were largely free of bulk motion artifacts (Fig. 3) and ADC maps were in agreement with myocardial diffusivities measured in volunteers. There was no significant difference between mean ADCs measured pre- and post-contrast (mean ADCPre=1.46±0.2x10-3mm2/s, ADCPost=1.58±0.3x10-3mm2/s, P=N.S.).


The volunteer study demonstrated that cDWI with CODE-M1M2 mitigated bulk motion artifacts and substantially increased the range of cardiac phases that can accommodate robust ADC measurement. While previous approaches which have required careful selection of the sequence timing and several repeated acquisitions[4], CODE-M1M2 was successful for all patients imaged using a single, predetermined 100ms ECG delay. These findings echo previous reports[3, 5, 6] of cDWI with M1M2 nulled encoding. Myocardial ADC values (1.4 to 1.6x10-3mm2/s) were also in agreement with these reports. The agreement in ADC values between pre- and post-contrast imaging highlights the relatively weak T1 dependence of the sequence for characterizing myocardial microstructure, which is important for the interpretation of ADC values in patients that may not receive contrast.


CODE-M1M2 cDWI significantly improved robustness to cardiac bulk motion compared to MONO cDWI. CODE-M1M2 cDWI also permits first and second moment nulling with a shorter TE than existing MOCO cDWI methods.


This research was supported by Siemens Healthcare, the Department of Radiological Sciences at UCLA and the Graduate Program in Biosciences at UCLA.


1. Gansevoort RT, Correa-Rotter R, Hemmelgarn BR, Jafar TH, Heerspink HJ, Mann JF, et al. Chronic kidney disease and cardiovascular risk: epidemiology, mechanisms, and prevention. Lancet. 2013;382(9889):339-52.

2. Nguyen C, Fan Z, Sharif B, He Y, Dharmakumar R, Berman DS, et al. In vivo three-dimensional high resolution cardiac diffusion-weighted MRI: A motion compensated diffusion-prepared balanced steady-state free precession approach. MRM. 2013;72(5):1257-67.

3. Stoeck CT, von Deuster C, Genet M, Atkinson D, Kozerke S. Second-order motion-compensated spin echo diffusion tensor imaging of the human heart. MRM. 2015; doi: 10.1002/mrm.25784

4. Rapacchi S, Wen H, Viallon M, Grenier D, Kellman P, Croisille P, et al. Low b-value diffusion-weighted cardiac magnetic resonance imaging: initial results in humans using an optimal time-window imaging approach. Investigative radiology. 2011;46(12):751-8.

5. Nguyen C, Fan Z, Sharif B, He Y, Dharmakumar R, Berman DS, et al. In vivo three-dimensional high resolution cardiac diffusion-weighted MRI: A motion compensated diffusion-prepared balanced steady-state free precession approach. MRM. 2013.

6. Welsh C, Di Bella E, Hsu E. Higher-Order Motion-Compensation for In Vivo Cardiac Diffusion Tensor Imaging in Rats. IEEE transactions on medical imaging. 2015.


Figure 1: cDWI pulse sequence diagrams for (A) bulk-motion sensitive, monopolar (MONO) encoding; (B) modified bipolar motion compensated (MOCO) encoding and (C) Convex Optimized Diffusion Encoding encoding with M1 and M2 motion compensation (CODE-M1M2)(b=350s/mm2 and 1.5x1.5mm in-plane resolution) which eliminates sequence dead time and reduces TE compared to MOCO. This improves SNR by 49% assuming myocardial T2=45ms.

Figure 2: Diffusion weighted images (A) from a typical healthy volunteer acquired at eight different cardiac phases with MONO and CODE-M1M2 cDWI. Myocardial LV ADC values (B) (Mean±SD) and percentage±SD of motion corrupted (ADC>3.0x10-3mm2/s) voxels (C) are shown for MONO and CODE-M1M2 encoding across the ten volunteers. Dots in (B) and (C) represent mean values for individual subjects. CODE-M1M2 cDWI is less sensitive to bulk motion than MONO and is not as dependent on precise sequence timing.

Figure 3: Example diffusion weighted images (three directions) and ADC maps acquired in three patients post-contrast undergoing cardiac MRI using CODE-M1M2 cDWI during a systolic cardiac phase (100ms following QRS). The DWI demonstrated robustness to bulk motion which was reflected in the quality of the ADC maps (mean ADC was 1.58±0.3x10-3mm2/s across all five patients).

Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)