Evaluation of synthetic images derived from a neural network in pediatric brain magnetic resonance imaging.
Authors
Affiliations (6)
Affiliations (6)
- Cincinnati Children's Hospital Medical Center, University of Cincinnati Department of Radiology, 3333 Burnet Avenue, Cincinnati, OH 45229-3026, United States. Electronic address: [email protected].
- Philips Innovative Technologies, Röntgenstr. 24-26, 22335 Hamburg, Germany. Electronic address: [email protected].
- University of Cincinnati Department of Radiology, 222 Piedmont Ave Ste 1200, Cincinnati, OH 45219, United States. Electronic address: [email protected].
- Cincinnati Children's Hospital Medical Center, 3333 Burnet Avenue, Cincinnati, OH 45229-3026, United States. Electronic address: [email protected].
- Cincinnati Children's Hospital Medical Center, 3333 Burnet Avenue, Cincinnati, OH 45229-3026, United States. Electronic address: [email protected].
- Philips Innovative Technologies, Röntgenstr. 24-26, 22335 Hamburg, Germany. Electronic address: [email protected].
Abstract
Synthetic MRI (SyMRI) is a technique used to estimate tissue properties and generate multiple MR sequence contrasts from a single acquisition. However, image quality can be suboptimal. To evaluate a neural network approach using artificial intelligence-based direct contrast synthesis (AI-DCS) of the multi-contrast weighted images to improve image quality. This prospective, IRB approved study enrolled 50 pediatric patients undergoing clinical brain MRI. In addition to the standard of care (SOC) clinical protocol, 2D multi-delay multi-echo (MDME) sequence was obtained. SOC 3D T1-weighted (T1W), 2D T2-weighted (T2W) and 2D T2W fluid-attenuated inversion recovery (FLAIR) images from 35 patients were used to train a neural network generating synthetic T1W, T2W, and FLAIR images. Quantitative analysis of grey matter (GM) and white matter (WM) apparent signal to noise (aSNR) and grey-white matter (GWM) apparent contrast to noise (aCNR) ratios was performed. 8 patients were evaluated. When compared to SyMRI, T1W AI-DCS had better overall image quality, reduced noise/artifacts, and better subjective SNR in 100 % (16/16) of evaluations. When compared to SyMRI, T2W AI-DCS overall image quality and diagnostic confidence was better in 93.8 % (15/16) and 87.5 % (14/16) of evaluations, respectively. When compared to SyMRI, FLAIR AI-DCS was better in 93.8 % (15/16) of evaluations in overall image quality and in 100 % (16/16) of evaluations for noise/artifacts and subjective SNR. Quantitative analysis revealed higher WM aSNR compared with SyMRI (p < 0.05) for T1W, T2W and FLAIR. AI-DCS demonstrates better overall image quality than SyMRI on T1W, T2W and FLAIR.