An Unsupervised Deep Learning Approach for Dynamic‐Exponential Intravoxel Incoherent Motion MRI Modeling and Parameter Estimation in the Liver

Enhanced parameter estimation in multiparametric arterial spin labeling using artificial neural networks

PURPOSE

Multiparametric arterial spin labeling (MP-ASL) can quantify cerebral blood flow (CBF) and arterial cerebral blood volume (CBVa). However, its accuracy is compromised owing to its intrinsically low SNR, necessitating complex and time-consuming parameter estimation. Deep neural networks (DNNs) offer a solution to these limitations. Therefore, we aimed to develop simulation-based DNNs for MP-ASL and compared the performance of a supervised DNN (DNNSup), physics-informed unsupervised DNN (DNNUns), and the conventional lookup table method (LUT) using simulation and in vivo data.

METHODS

MP-ASL was performed twice during resting state and once during the breath-holding task. First, the accuracy and noise immunity were evaluated in the first resting state. Second, CBF and CBVa values were statistically compared between the first resting state and the breath-holding task using the Wilcoxon signed-rank test and Cliff's delta. Finally, reproducibility of the two resting states was assessed.

RESULTS

Simulation and first resting-state analyses demonstrated that DNNSup had higher accuracy, noise immunity, and a six-fold faster computation time than LUT. Furthermore, all methods detected task-induced CBF and CBVa elevations, with the effect size being larger with the DNNSup (CBF, p = 0.055, Δ = 0.286; CBVa, p = 0.008, Δ = 0.964) and DNNUns (CBF, p = 0.039, Δ = 0.286; CBVa, p = 0.008, Δ = 1.000) than that with LUT (CBF, p = 0.109, Δ = 0.214; CBVa, p = 0.008, Δ = 0.929). Moreover, all the methods exhibited comparable and satisfactory reproducibility.

CONCLUSION

DNNSup outperforms DNNUns and LUT with respect to estimation performance and computation time.

Estimation of Cerebral Blood Flow and Arterial Transit Time From Multi-Delay Arterial Spin Labeling MRI Using a Simulation-Based Supervised Deep Neural Network

BACKGROUND

An inherently poor signal-to-noise ratio (SNR) causes inaccuracy and less precision in cerebral blood flow (CBF) and arterial transit time (ATT) when using arterial spin labeling (ASL). Deep neural network (DNN)-based parameter estimation can solve these problems.

PURPOSE

To reduce the effects of Rician noise on ASL parameter estimation and compute unbiased CBF and ATT using simulation-based supervised DNNs.

STUDY TYPE

Retrospective.

POPULATION

One million simulation test data points, 17 healthy volunteers (five women and 12 men, 33.2 ± 14.6 years of age), and one patient with moyamoya disease.

FIELD STRENGTH/SEQUENCE

3.0 T/Hadamard-encoded pseudo-continuous ASL with a three-dimensional fast spin-echo stack of spirals.

ASSESSMENT

Performances of DNN and conventional methods were compared. For test data, the normalized mean absolute error (NMAE) and normalized root mean squared error (NRMSE) between the ground truth and predicted values were evaluated. For in vivo data, baseline CBF and ATT and their relative changes with respect to SNR using artificial noise-added images were assessed.

STATISTICAL TESTS

One-way analysis of variance with post-hoc Tukey's multiple comparison test, paired t-test, and the Bland-Altman graphical analysis. Statistical significance was defined as P < 0.05.

RESULTS

For both CBF and ATT, NMAE and NRMSE were lower with DNN than with the conventional method. The baseline values were significantly smaller with DNN than with the conventional method (CBF in gray matter, 66 ± 10 vs. 71 ± 12 mL/100 g/min; white matter, 45 ± 6 vs. 46 ± 7 mL/100 g/min; ATT in gray matter, 1424 ± 201 vs. 1471 ± 154 msec). CBF and ATT increased with decreasing SNR; however, their change rates were smaller with DNN than were those with the conventional method. Higher CBF in the prolonged ATT region and clearer contrast in ATT were identified by DNN in a clinical case.

DATA CONCLUSION

DNN outperformed the conventional method in terms of accuracy, precision, and noise immunity.

EVIDENCE LEVEL

3 Technical Efficacy: Stage 1.