Post-contrast 3D T1 weighted sequences (T1WI) showed excellent soft tissue contrast and contrast enhancing effect after contrast injection and widely used in brain MRI for evaluation of pathologies. However, high-resolution isotropic T1WI usually require long scan time and might increase motion artifact and patient anxiety, which can compromise image quality. 

Parallel imaging is the most commonly used acceleration method for decreasing the scan time with reducing the k-space line. However, parallel imaging with a high acceleration factor usually results in increased image noise with an inhomogeneous noise distribution compared to standard images. This phenomenon was more pronounced with an increase in the acceleration factor. These alternative accelerated MRI techniques have trade-off in image quality with decreases the SNR.

Recently developed deep learning techniques have been widely applied to improve image quality with decreasing the image noise. Single image super-resolution is a deep learning‐based nonlinear interpolation technique and it denoises the images, enhance edges, increase sharpness, and address Gibbs ringing using a neural network. It is able to improve spatial resolution of thin-sliced MRI without compromising SNR or requiring additional MRI hardware or scan time and showed superior performance for generating thin-sliced MRI compared with commonly used techniques. The purpose of our study was to evaluate the diagnostic performance of Deep learning based super-resolution in iterative denoising of post-contrast T1-weighted VIBE (DL-VIBE) for evaluating intracranial enhancing lesions compared with standard 3D-MPRAGE. To our knowledge, deep learning based reconstruction of contrast-enhanced VIBE (DL-VIBE) for evaluating intracranial enhancing lesions has not been previously performed.

We consecutively peformed post-contrast 3D-MPRAGE (acquisition time 5 min 32 s) and DL-VIBE (1 min 49 s)  during same image session in 97 patients (59.2±15.3 years; range, 20-89 years) for evaluation of intracranial enhancing lesions. Two neuroradiologists evaluated the overall image quality, gray-white matter differentiation, flow-related artifact, conspicuity of enhancing lesions and the number and contrast-to-noise ratio (CNR) of the enhancing lesions from the two different sequences.