Assessment of diastolic function using cine-CMR is limited by its temporal resolution and may yield erroneous left ventricular [LV] diastolic filling parameters. Recently, an optimized data acquisition/reconstruction hybrid strategy that uses radial sampling, called Modulo-Prime Spokes [MoPS] [1], was shown to achieve temporal frame rates comparable to echocardiograpy within a single breath-hold per slice at 15ms resolution, or 66 frames/sec (66Hz). In this study, we evaluate whether this novel high-temporal resolution cine-CMR approach can result in improved assessment of diastolic function using time-volume curves when compared to conventional Cartesian cine-CMR.

Imaging Protocol: 11 healthy volunteers (34±7 years) were imaged at 1.5T (Philips Achieva) using short-axis balanced-SSFP cine-CMR (6mm slice thickness and 4mm gap, 15 slices for whole-heart coverage, 32 cardiac phases) under expiratory breath-holds. This sequence was immediately followed by the MoPS-Cine acquisition on the same geometric prescription (8mm slices, 2mm gap), which acquired a period greater than 1R-R interval (~1.4 sec) over a 12 heart-beat breath-hold (ie. 6x2R-R ECG-triggered acquisition).

MoPS Acquisition and Reconstruction: A combined 'Optimized MoPS' strategy (180 uniformly distributed spokes, with MoPS prime = 181; rotation angle = 22° ≈ Golden Angle / 5; nTFE = 30; TR=2.5 ms; FA = 60°; reconstructed spatial resolution = 0.875x0.875 mm²; slice thickness = 8mm with slice gap = 2mm) was used (Fig 1A), allowing GPU-accelerated reconstruction using CUDA-PTX (NVIDIA, Santa Clara, CA) at a temporal resolution of 6 consecutive TRs, or 15ms, for each breath-held slice with ~90 cardiac phases in ~10 seconds [1]. Each image was undersampled at 20% density (36/180 spokes), and the reconstructed cardiac cycle spanned 1.4 seconds from the reference ECG trigger. No iterative reconstruction approach was used, and the undersampled streaking was removed by employing a radial UNFOLD-based low-pass filter [1,2], whose characteristic streak frequencies were mathematically optimized to be filterable without removing physiologic information by using a specific set of acquisition/recon parameters. The reference cine-CMR sequence employed SENSE parallel imaging (R=2), 6mm thickness and gap = 4 mm, and a lower spatial resolution at 1.375x1.375 mm². A custom workstation hardware was used to perform simultaneous reconstruction and LV segmentation within the clinical workflow.

Clinical Post-Processing: Diastolic function parameters were calculated offline using an operator-assisted LV analysis processing tool in MATLAB (The Mathworks, Natick MA). This tool incorporated operator control with a level-set based LV segmentation algorithm [3] that performed a small number of iterations (n=10) for each image. The operator rapidly delineated the endocardial contours across all cardiac phases from the basal to apical LV slices by allowing the contours to converge towards the endocardial border in real-time (Fig. 1B). The console was customized to blind the operator from any volume-derived measurements during segmentation. Volume vs time curves were generated for each LV.

Analysis: Systolic and diastolic CMR parameters [4] were generated from the derived time-volume curves (Fig 1C). These include: End Systolic Volume [ESV]; End Diastolic Volume [EDV]; Ejection Fraction [EF]; Diastolic phase duration [Ddur]; normalized Peak Flow Rate [nPFR] - (Δvolume/Δtime) normalized by the R-R interval; normalized Time to Peak Flow Rate [nTPFR] - percentage of R-R cycle between end-systole and peak flow; and Diastolic Volume Recovery [DVR] - normalized time fraction of Ddur at which 75% stroke volume recovery occurs.

All acquisition, reconstruction, and clinical post-processing were completed successfully. The average processing time was 0.51±0.02 sec (MoPS data with 400x400 voxels/image) vs 0.23±0.02 sec (Reference data with 240x240 voxels/image) per operator interaction on each image, reported as an average of 250 computations. Table 1 shows the analysis results. No difference was seen in the LV systolic parameters (p=NS). All diastolic assessment parameters that are dependent on temporal frame rate were significantly higher in the high-resolution MoPS-cine approach. Namely, Ddur = 69±4 vs 66±3% (MoPS vs Cartesian; p<0.05); nPFR = 375±128 vs 327±113 mL/sec/RRdur (p<0.05); and nTPFR = 18.2±2.7 vs 16.6±2.8% (p<0.05). DVR, which inherently interpolates between cardiac phases upon its calculation, was non-significant (P=NS). The improved temporal resolution in the MoPS-cine derived time-volume curves yielded smaller quantization error, likely leading to improved precision of diastolic assessment parameters that are sensitive to higher temporal frame rates.

We demonstrate the feasibility of a novel high-temporal cine-CMR approach for assessment of LV systolic and diastolic functions. The MoPS-cine-derived time-volume curves with a higher temporal resolution yielded potentially more precise diastolic assessment parameters compared to the conventional cine-CMR approach. While there was an approximately three-fold increase in the number of cardiac phases to delineate, the adaptation of an existing LV segmentation algorithm for easy-to-use operator-assistance in contour delineation step made the post-processing not cumbersome.