Super-fast 3T Prostate MRI Using High Gradient Strength and Deep Learning: Initial Experience

Multiparametric MRI (mpMRI) of the prostate has become the most important non-invasive diagnostic tool for assessment of prostate cancer and is the baseline for MRI targeted biopsy (1,2). As the incidence of prostate cancer is high with an estimated 290,000 new cases for 2023 in the United States alone (3), the need for widespread provision of prostate mpMRI is immense. However, current clinical MRI-protocols are long with acquisition times of >30 minutes, potentially limiting the number of examined patients. According to the guidelines of the Prostate Imaging Reporting and Data System (PI-RADS) a sufficient mpMRI protocol must include diffusion weighted imaging (DWI), T2-weighted (T2w) imaging, dynamic contrast enhanced imaging and T1-weighted imaging pre and post administration of contrast (4,5). Different approaches have been developed to shorten the protocol itself or to accelerate acquisition times. For instance, a significant reduction of T2w-sequence acquisition times was achieved by employing deep learning methods (6) or advancements of compressed sensing (7), while at the same time images had improved quality. Different studies showed equal performance of biparametric MRI (bpMRI) protocols compared to the standard multiparametric protocol, effectively reducing the acquisition time down to 5 minutes (8,9,10). This was done by focusing only on DWI and T2w imaging while omitting the dynamic contrast enhanced sequence and T1-weighted sequences, as the additional diagnostic value of these is supposed to be limited (11).

Recent developments in MRI techniques allow for ultra-high diffusion gradient strengths of up to 500 mT/m and slew rates of up to 600 T/m/s, thus reducing echo times and acquisition times by faster establishment of the diffusion gradient (12,13). Furthermore, these gradients are able to image at small scales with a high signal-to-noise ratio, consecutively enhancing sensitivity for detection of tissue microstructures (14,15). Due to the experimental nature of these gradients, this technique has only been investigated in a research setting in healthy volunteers (16), but not in a real world clinical setting, let alone in prostate imaging.

Therefore, the aim of the study was to establish a super-fast abbreviated bpMRI protocol for patients with suspicion for prostate cancer using both ultra-high gradients and deep learning reconstruction for DWI- and T2w-sequences. Besides the assessment of acquisition times, the main objective of this study was to assess the overall image quality of bpMRI and mpMRI and the influence on PI-RADS scores.