This study was approved by the regional research ethics committees and all participants gave written informed consent. CMR was performed at 3T (Siemens Skyra, Erlangen, Germany) using an anterior body and spine receive coil array.
We modified a cine DENSE sequence with a spiral k-space trajectory  by reducing the imaged field of view (FOV) to a small square region around the left ventricle while avoiding aliasing artefacts (Fig. 1). This was achieved via perpendicular in-plane slice selection gradients applied to the first and second of the three radiofrequency (RF) pulses required to produce the stimulated echo. The smaller FOV means that fewer spiral interleaves can be used for a given spatial resolution without undersampling. To provide a satisfactory excitation profile, with minimal additional echo time, a time-reversed pair of asymmetric RF pulses (81% and 19% asymmetry) was used.
The in-plane excitation profile and signal-to-noise ratio (SNR) of the proposed method was assessed in an 11 cm diameter cylindrical phantom filled with 40 g/L agar in tap water. T1 and T2 were measured at 1060 ± 20 ms and 58.1 ± 0.7 ms, respectively.
The in-plane excitation profile was assessed by acquiring DENSE data with a stimulated echo FOV of 50 × 50 mm2 (determined by the slice selection of the first two RF pulses in Fig. 1) and a larger readout FOV of 250 × 250 mm2.
To determine the maximum size of the stimulated echo FOV for a given readout FOV without aliasing artefacts DENSE imaging was performed using a 70 × 70 mm2 readout FOV and a stimulated echo FOV of 70 × 70 mm2, 90 × 90 mm2 and 110 × 110 mm2.
To determine the effect of the reduced FOV technique on the magnitude image SNR, DENSE acquisitions were performed with 4 protocols, each repeated 20 times to provide noise estimates . Three protocols were those used in the pilot healthy volunteer study below (20RR intervals, 360 × 360 mm2 FOV; 14RR-intervals, 224 × 224 mm2 FOV; and 8RR-intervals, 120 × 120 mm2 FOV) and the fourth was identical to the 20RR-interval acquisition without the reduced FOV technique. Single image SNR was calculated using the multiple repetitions method  and a mean ± standard deviation was calculated for an artefact-free region within the phantom.
A pilot study (originally reported in ) evaluated the performance of this novel sequence in vivo with increasingly short breath hold durations. Eight healthy volunteers were imaged using the novel reduced FOV sequence with three different FOVs and breath hold durations.
DENSE was performed in a mid-ventricular short-axis slice using 3.2 × 3.2 mm2 acquired spatial resolution, 8 mm slice thickness, 30 ms temporal resolution, 2 spiral interleaves per RR-interval, 2-directional encoding at 0.06 cycles/mm, chemical shift selective fat saturation, repetition time/echo time (TR/TE) 15/1.0 ms, and variable flip angle (maximum of 20°). Unwanted echo pathways were minimised via CSPAMM-like encoding and through-plane dephasing (0.08 cycles/mm). Second-order B0 shimming was performed using a cardiac specific method . As in the phantom SNR study, the three protocols were run with FOVs of 360 × 360 mm2, 224 × 224 mm2 and 120 × 120 mm2 (equal stimulated echo and readout FOV) with corresponding breath hold durations of 20RR intervals, 14RR intervals, and 8RR intervals. The reduction in FOV was achieved via a reduction in the total number of spiral interleaves from 8, to 6, to 4.
Healthy control cohort
The healthy control cohort (n = 18) comprised individuals who did not have a history of medical illness, were not taking regular medication, and did not have evidence of cardiac structural or functional impairment on CMR.
The DCM cohort was prospectively recruited (n = 29). Inclusion criteria were meeting criteria for a diagnosis of DCM (Supplementary materials), over 18 years of age, sinus rhythm, and the absence of a contraindication to CMR.
All participants underwent CMR for assessment of cardiac structure, function, and myocardial strain.
Acquisition and assessment of cardiac structure and function
End-expiratory breath hold balanced steady-state free precession (bSSFP) cine images were acquired in the 3 long-axis planes (horizontal long axis, HLA; right ventricular outflow tract, RVOT; left ventricular outflow tract, LVOT); and 8 mm short-axis slices (2 mm gap) from the atrioventricular ring to the apex as previously described  (acquisition parameters in Supplementary materials). Biventricular volumes, function, and left ventricular (LV) mass were measured using a semi-automated threshold-based technique (CMRtools, Cardiovascular Imaging Solutions, London, UK). Left ventricular wall thickness and left atrial area were assessed as outlined in Supplementary materials. All volume and mass measurements were indexed to body surface area and referenced to age and gender .
Acquisition and assessment of myocardial strain data using modified cine spiral DENSE sequence
Based on the initial pilot study, DENSE acquisitions were performed using the 14RR interval protocol (224 × 224 mm2 FOV, readout and stimulated echo FOV equal) described above. Images were acquired at the mid-ventricular short-axis (SAX) level and in 2 long-axis planes (horizontal, HLA and vertical, VLA) using the novel reduced FOV sequence and an optimised protocol. All images were reconstructed online at the scanner. Comparative images were not acquired with a breath hold duration of 20 RR intervals, as these data were acquired within a longer protocol. We did not wish to subject patients with dilated cardiomyopathy to multiple long breath holds that they would find difficult and potentially lead to aborted studies.
Images were analysed and strain extracted from the DENSE data using semi-automated MATLAB (The Mathworks, Natick, MA) post-processing software from the University of Virginia [19,20,21]. The first stage of analysis was anatomical delineation, using either a contour or a region of interest covering the LV in the imaged slice. For long-axis images, this was done with a single line contour placed in the mesocardium, midway between the epicardium and endocardium (Fig. 2). For SAX images, both contour analysis and a region of interest were defined, with the region of interest manually defined between endo- and epicardial borders (Fig. 2). These contours were defined in either the peak systolic frame or a diastolic frame before automated propagation to the other frames (motion guided segmentation ), with subsequent manual adjustment as required. The frames with poor SNR at the end of cardiac cycle, where the myocardial outline could not be discerned, were discarded. Strain was then calculated in the segmented areas, generating regional polar-strain/time curves for radial and circumferential strain, contour strain/time curves for longitudinal strain in 2 planes and contour strain/time curves for short-axis strain. From these, global peak strain results were derived for each participant: SAX contour strain, longitudinal strain in the HLA and VLA planes, radial strain, and circumferential strain.
Continuous data are expressed as mean ± standard deviation or median (inter-quartile range) and compared using the t test or Mann–Whitney test, respectively. Categorical data are expressed as number and percentages, and compared using Fisher’s exact test. Univariable linear regression was used to display the relationship between strain data and LVEF. Correlation testing using the Pearson correlation coefficient was performed to define the relationship between strain data and CMR parameters and visualised using the corrplot package in R. A random subset of studies (n = 5 controls and n = 5 DCM) was analysed by an independent operator and intraclass correlation coefficients were calculated to evaluate inter-observer variability. A p value of < 0.05 was considered significant and all analyses were conducted in the R statistical environment (version 3.3.1) or MATLAB (R2018a, pilot data only).