Cardiac magnetic resonance (CMR) is recognized as the reference standard non-invasive imaging technique for the evaluation of cardiac chambers volume and systolic function, for both ischaemic and non-ischaemic diseases [1], [2], [3]. Furthermore, this technique delivers unmatched native and post-contrast myocardial tissue characterisation with qualitative and quantitative techniques, with weighted images, mapping sequences and late gadolinium enhancement (LGE) images [4], [5], [6]. However, several limitations often reduce prescription and performance of CMR exams, such as claustrophobia, presence of medical devices not approved for magnetic field environment, and the overall exam duration, especially when detailed tissue characterisation is required in patients with reduced environment tolerance or impaired clinical status.

According to specific reason of patient referral and underlining clinical question, CMR scan typically involves the acquisition of multiple sequences that almost always include the assessment of cardiac volumes and function, and qualitative tissue characterisation. In order to perform morpho-functional evaluation, a set of three left ventricle (LV) long-axis views (2-, 3- and 4-chamber views) and a short-axis stack (approximately 12–15 slices) are commonly acquired, maintaining breath-holds of about 8–12 s each, depending on heart rate, for a total duration of several minutes. During this time the patient, particularly if compromised or fragile, may become fatigued or unable to comply with the breath-hold instructions, thus impairing image quality and sometimes overall diagnostic accuracy. Moreover, in addition to cardiac functional evaluation, there is increasing demand from clinicians for tissue characterisation, mostly for the detection of myocardial scar/fibrosis, and such assessment is commonly performed with two-dimensional (2D) single segment breath-hold acquisitions, in order to mitigate respiratory motion.

Thanks to technical development, new sequences designed to obtain cine images in less time and breath-holds with high image quality have been recently developed. In detail, the acquisition design of these new sequences intentionally samples only a fraction of k-space to substantially shorten scan time, while a deep learning-based reconstruction algorithm, trained on fully sampled reference data, learns to estimate the unacquired k-space components and reconstruct a complete and accurate image [7], [8], [9]. Moreover, 2D-multisegment phase sensitive LGE (2D-MS PS LGE) sequences combined with deep learning-based reconstruction and noise reduction algorithm are already available, providing superior image quality with similar diagnostic accuracy and a significantly shortened time of acquisition compared to standard 2D single segmented LGE sequences (2D-SS PS LGE) [10].

The aim of this study is to compare ‘standard protocol’ composed by conventional two-dimensional (2D) balanced steady state free precession cine sequences (ASSET bSSFP, or ASSET cine) and standard 2D single segmented phase sensitive inversion recovery LGE sequences (2D-SS PS LGE) versus a ‘fast protocol’ composed by vendor-specific accelerated 2D bSSFP cine sequence with deep learning reconstruction (Sonic DL bSSFP, or Sonic DL cine) and 2D-multisegment phase sensitive inversion recovery LGE sequences (2D-MS PS LGE) in terms of image quality, biventricular functional assessment and post-contrast myocardial evaluation.