Accelerating CEST imaging using a model‐based deep neural network with synthetic training data

Abstract

PurposeTo develop a model‐based deep neural network for high‐quality image reconstruction of undersampled multi‐coil CEST data.Theory and MethodsInspired by the variational network (VN), the CEST image reconstruction equation is unrolled into a deep neural network (CEST‐VN) with a k‐space data‐sharing block that takes advantage of the inherent redundancy in adjacent CEST frames and 3D spatial–frequential convolution kernels that exploit correlations in the x‐ω domain. Additionally, a new pipeline based on multiple‐pool Bloch–McConnell simulations is devised to synthesize multi‐coil CEST data from publicly available anatomical MRI data. The proposed network is trained on simulated data with a CEST‐specific loss function that jointly measures the structural and CEST contrast. The performance of CEST‐VN was evaluated on four healthy volunteers and five brain tumor patients using retrospectively or prospectively undersampled data with various acceleration factors, and then compared with other conventional and state‐of‐the‐art reconstruction methods.ResultsThe proposed CEST‐VN method generated high‐quality CEST source images and amide proton transfer‐weighted maps in healthy and brain tumor subjects, consistently outperforming GRAPPA, blind compressed sensing, and the original VN. With the acceleration factors increasing from 3 to 6, CEST‐VN with the same hyperparameters yielded similar and accurate reconstruction without apparent loss of details or increase of artifacts. The ablation studies confirmed the effectiveness of the CEST‐specific loss function and data‐sharing block used.ConclusionsThe proposed CEST‐VN method can offer high‐quality CEST source images and amide proton transfer‐weighted maps from highly undersampled multi‐coil data by integrating the deep learning prior and multi‐coil sensitivity encoding model.

Deep‐learning‐based super‐resolution for accelerating chemical exchange saturation transfer MRI

“…To address this issue, one approach is to include neighboring-offset images in the network input when reconstructing each center CEST source image. 47 The other potential reason could be the various degrees of improvement in spatial resolution of the CEST source images across different offset frequencies, as evidenced by the PSNR and SSIM values (Figures 4B, 5C and 6). To generate the Z-spectrum, a pixel intensity at different offset frequencies is normalized by the pixel's intensity at M 0 before plotting the pixel intensity values as a function of offset frequencies.…”

“…Moreover, with its adaptability to any acquisition sequence, non-requirement of additional hardware such as array receiver coils and its superior performance, DLSR has been adopted for a wide range of clinical scans, including brain, 43,44 musculoskeletal, 32 and cardiac MRI. 45 However, using DLSR for CEST MRI has only been explored in small-scale research studies 46,47 due to a lack of the large CEST datasets required for network development.…”

Highly‐accelerated CEST MRI using frequency‐offset‐dependent k‐space sampling and deep‐learning reconstruction

“…These approaches may enable the acquisition of incomplete data with higher efficiency. Conventional iterative [29][30][31]34 (e.g., CS-SENSE, k−t Sparse SENSE) and deep-learning 41,42 based MRI acceleration approaches, have not fully utilized CEST-related priors in their reconstruction. Recently a k−ω ROSA approach 33 incorporates the symmetry and asymmetry properties of z-spectra and achieves 5× acceleration.…”

“…With recent advances in deep learning techniques for MRI reconstruction, [36][37][38][39][40] deep learning networks have been exploited to reconstruct de-aliased CEST images from under-sampled k-space data. 41,42 These networks have shown higher reconstruction accuracy compared to compressed sensing-based methods and/or parallel imaging. However, previous deep learning approaches for CEST acceleration have seldom incorporated CEST spectral priors directly into the network architecture.…”