Ultrafast brain MRI within 100 s based on deep learning reconstruction: a prospective feasibility study

Prolonged brain MRI examinations increase susceptibility to motion and may limit the feasibility of MRI in patients who cannot reliably cooperate. Ultrafast MRI may mitigate these challenges by substantially reducing acquisition time while preserving clinically useful image quality. This study aims to investigate the feasibility of ultrafast MRI, an EPI-based method, namely Deep-learning Enabled Precise Imaging via multi-Contrast multi-shoT EPI for Accelerated head scan (DEPICTA). In this prospective study, we included 124 consecutive patients with a clinical indication for brain MR underwent both conventional MRI and DEPICTA from August to September 2025. Two neuroradiologists independently assessed overall image quality, gray-white matter (GM-WM) differentiation, and artifacts. SNR, ventricular width, lesion size, and lesion ADC were also assessed. The T2*-weighted/SWI module was not included in the comparative image-quality or quantitative analyses. Among 124 participants (mean age, 49 ± 22 years; 64 male and 60 female), DEPICTA achieved an acquisition time of 1 min 27 s. The overall image quality and GM-WM difference of ultrafast MRI were lower than conventional MRI, but ultrafast MRI still demonstrated sufficient overall image quality. The ultrafast MRI had fewer artifacts and higher SNR on T1-FLAIR and DWI images compared with the conventional MRI ( P < 0.05). Ultrafast MRI showed good consistency with conventional MRI in terms of lesion size, lesion ADC, and width of the lateral and third ventricles. The width of the fourth ventricle showed differences ( P = 0.006). DEPICTA enables multi-contrast brain MRI within 100 s and demonstrates promising image-quality and quantitative-measurement comparability with conventional MRI. These findings support its potential and feasibility as an ultrafast brain MRI approach, particularly when examination time is critical.