Rapid framework for quantitative magnetization transfer imaging with interslice magnetization transfer and dictionary‐driven fitting approaches

Abstract

Purpose:To develop a rapid framework for quantitative magnetization transfer (qMT) imaging based on the 2D interslice MT and dictionary-driven fitting approaches. Methods: Sequential balanced steady-state free precession (bSSFP) scanning was performed on a whole human brain in a total of 12 conditions from six different interslice gaps and two different flip angles. To obtain qMT maps, the acquired 12 datasets were fitted to a dictionary predefined by using Bloch equation simulations based on the two-pool MT model. The proposed qMT method was compared to the conventional qMT methods, in terms of qMT parameter maps and processing time. Results: The proposed method yielded qMT maps similar to those of the conventional method, indicating feasibility of modulating MT saturation frequency and power with the interslice gap and flip angle. The whole-brain qMT imaging could be completed in 8 min because of the absence of separate MT pulses. The time for processing qMT parameters was significantly reduced through the dictionary-driven approach; it is 1000 times shorter than that without the dictionary-driven approach and 3 times shorter than that with the spoiled gradient recalled echo-qMT method that uses an analytical solution. Conclusion: The proposed dictionary-driven interslice qMT imaging provides qMT maps close to those from the conventional method with significantly reduced scan time and postprocessing time, which can make qMT imaging more clinically acceptable. K E Y W O R D Sbalanced steady-state free precession, dictionary analysis, interslice, magnetization transfer, quantitative imaging

qMTNet: Accelerated quantitative magnetization transfer imaging with artificial neural networks

“…As a trade‐off for the lower‐fitting quality, qMTNet‐2 and qMTNet‐1 showed the highest acceleration factor out of the four methods utilizing ANN. Compared with the dictionary‐driven framework in Kim et al, 14 both qMTNet achieved an acquisition acceleration factor of 3 from the reduced number of acquired data. qMTNet‐acq takes 0.009 s on average to produce one set of eight MT images, which is negligible compared with the acquisition time.…”

“…Imaging experiments were conducted on seven healthy volunteers (five males, two females, 24‐29 years old) using a Siemens 3T Tim Trio system (Siemens Medical Solutions, Erlangen, Germany) for developing networks, and on four healthy young subjects (three men, one woman, 25‐27 years old) and four healthy older subjects (three men, one woman, 61‐76 years old) using a Siemens Verio 3T scanner for testing the generalizability of the networks. MT data were acquired with both conventional 5 and interslice qMT imaging methods, following the procedures in Kim et al 14 The overall acquisition chronogram is shown in Supporting Information Figure . Conventional qMT data were acquired with a train of off‐resonance RF pulses, which was applied for 3 s without a selective gradient.…”

“…To construct the ground truth qMT parameters for training the networks, the dictionary‐driven fitting method was used, as described in detail in Kim et al 14 Briefly, a dictionary of MR signal intensities was constructed based on the tissue properties through Bloch equation simulation incorporating the two‐pool MT model. The dictionary contained seven variables: T 1 ; T 2 values of the free pool; flip angle; off‐resonance frequency; two qMT parameters, pool fraction ( F ) and magnetization exchange rate ( k f ); and the MR intensities.…”

“…The acquired MT data are compared with entries in the dictionary to fit the qMT parameters to the data, fundamentally similar to MR fingerprinting 24 . Interslice qMT imaging with dictionary‐driven fitting took 19.2 s to acquire one slice and 0.17 s to fit one pixel, significantly faster than previous methods 14 …”

“…Recently, a framework was proposed to combine interslice MT acquisition with dictionary‐driven fitting to alleviate these problems 14 . Interslice MT utilizes the incidental MT effect 15‐19 in sequential multislice scanning.…”