Deep‐Learning‐Based Contrast Synthesis From MRF Parameter Maps in the Knee Joint

Abstract

BackgroundMagnetic resonance fingerprinting (MRF) is a method to speed up acquisition of quantitative MRI data. However, MRF does not usually produce contrast‐weighted images that are required by radiologists, limiting reachable total scan time improvement. Contrast synthesis from MRF could significantly decrease the imaging time.PurposeTo improve clinical utility of MRF by synthesizing contrast‐weighted MR images from the quantitative data provided by MRF, using U‐nets that were trained for the synthesis task utilizing L1‐ and perceptual loss functions, and their combinations.Study TypeRetrospective.PopulationKnee joint MRI data from 184 subjects from Northern Finland 1986 Birth Cohort (ages 33–35, gender distribution not available).Field Strength and SequenceA 3 T, multislice‐MRF, proton density (PD)‐weighted 3D‐SPACE (sampling perfection with application optimized contrasts using different flip angle evolution), fat‐saturated T2‐weighted 3D‐space, water‐excited double echo steady state (DESS).AssessmentData were divided into training, validation, test, and radiologist's assessment sets in the following way: 136 subjects to training, 3 for validation, 3 for testing, and 42 for radiologist's assessment. The synthetic and target images were evaluated using 5‐point Likert scale by two musculoskeletal radiologists blinded and with quantitative error metrics.Statistical TestsFriedman's test accompanied with post hoc Wilcoxon signed‐rank test and intraclass correlation coefficient. The statistical cutoff P <0.05 adjusted by Bonferroni correction as necessary was utilized.ResultsThe networks trained in the study could synthesize conventional images with high image quality (Likert scores 3–4 on a 5‐point scale). Qualitatively, the best synthetic images were produced with combination of L1‐ and perceptual loss functions and perceptual loss alone, while L1‐loss alone led to significantly poorer image quality (Likert scores below 3). The interreader and intrareader agreement were high (0.80 and 0.92, respectively) and significant. However, quantitative image quality metrics indicated best performance for the pure L1‐loss.Data ConclusionSynthesizing high‐quality contrast‐weighted images from MRF data using deep learning is feasible. However, more studies are needed to validate the diagnostic accuracy of these synthetic images.Evidence Level4.Technical EfficacyStage 1.

Emerging Trends in Magnetic Resonance Fingerprinting for Quantitative Biomedical Imaging Applications: A Review

“…To make MRF clinically relevant for radiologists, we need to incorporate mechanisms to reconstruct fully sampled contrast-weighted images like -weighted, -weighted, and FLAIR in the MRF reconstruction pipeline. In [ 73 , 74 ], the authors propose synthetic contrast generation from MRF data using deep learning methods.…”

Editorial for “Deep‐Learning‐Based Contrast Synthesis From MRF Parameter Maps in the Knee Joint”

“…In the study 9 we are editorializing, they propose to utilize qMRI parameters acquired by MRF (proton density, T 1 relaxation time, T 2 relaxation time, and B1+ field) to synthesize various contrast‐weighted MR images that are acquired during routine clinical MRI scans of the knee joint. This was accomplished by training DCNNs, more specifically U‐Nets to synthesize these contrasts using MRF‐acquired quantitative parameter maps as inputs to the neural network.…”

“…In this current study, 9 MRI data from 184 knee joints of subjects were acquired at a 3 T MR scanner. Conventional contrast‐weighted images were collected using three‐dimensional (3D) sequences (proton density (PD)‐weighted sampling‐perfection‐with‐application‐optimized‐contrasts‐using‐different‐angle‐evaluation (SPACE) sequence, fat‐saturated T 2 ‐weighted SPACE sequence, and water excitation double‐echo‐steady‐state (DESS) sequence).…”

“…7 Some of these problems may be overcome by utilizing deep-convolutional neural networks (DCNNs) in the synthesis instead of modelbased synthesis. 7 Previously, DCNNs have been successfully utilized for many different image synthesis tasks in medical imaging, including cross-modality synthesis between MRI and CT. 8 In the study 9 we are editorializing, they propose to utilize qMRI parameters acquired by MRF (proton density, T 1 relaxation time, T 2 relaxation time, and B1+ field) to synthesize various contrast-weighted MR images that are acquired during routine clinical MRI scans of the knee joint. This was accomplished by training DCNNs, more specifically U-Nets to synthesize these contrasts using MRF-acquired quantitative parameter maps as inputs to the neural network.…”