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

Multiparametric arterial spin labeling (MP-ASL) can quantify cerebral blood flow (CBF) and arterial cerebral blood volume (CBV_a). 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 (DNN_Sup), physics-informed unsupervised DNN (DNN_Uns), and the conventional lookup table method (LUT) using simulation and in vivo data.

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 CBV_a 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.

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

DNN_Sup outperforms DNN_Uns and LUT with respect to estimation performance and computation time.