CEST MR fingerprinting (CEST-MRF) for brain tumor quantification using EPI readout and deep learning reconstruction

Abstract

Purpose: To develop a clinical CEST MR fingerprinting (CEST-MRF) method for brain tumor quantification using EPI acquisition and deep learning reconstruction.

Methods: A CEST-MRF pulse sequence originally designed for animal imaging was modified to conform to hardware limits on clinical scanners while keeping scan time under 2 min. Quantitative MRF reconstruction was performed using a deep reconstruction network (DRONE) to yield the water relaxation and chemical exchange parameters. The feasibility of the six parameter DRONE reconstruction was tested in simulations using a digital brain phantom. A healthy subject was scanned with the CEST-MRF sequence, conventional MRF and CEST sequences for comparison. Reproducibility was assessed via test-retest experiments and the concordance correlation coefficient calculated for white matter and gray matter. The clinical utility of CEST-MRF was demonstrated on four patients with brain metastases in comparison to standard clinical imaging sequences. Tumors were segmented into edema, solid core, and necrotic core regions and the CEST-MRF values compared to the contra-lateral side.

Results: DRONE reconstruction of the digital phantom yielded a normalized RMS error of ≤7% for all parameters. The CEST-MRF parameters were in good agreement with those from conventional MRF and CEST sequences and previous studies. The mean concordance correlation coefficient for all six parameters was 0.98 ± 0.01 in white matter and 0.98 ± 0.02 in gray matter. The CEST-MRF values in nearly all tumor regions were significantly different (P = 0.05) from each other and the contra-lateral side.

Conclusion: Combination of EPI readout and deep learning reconstruction enabled fast, accurate and reproducible CEST-MRF in brain tumors.

Keywords: DRONE; chemical exchange rate; chemical exchange saturation transfer (CEST); deep learning; magnetic resonance fingerprinting (MRF); pH.

FIGURE 1

(A) Clinical CEST‐MRF pulse sequence shown for one schedule point. The magnetization is saturated with a Gaussian‐shaped pulse train and exchanges with the water. The water signal is then excited and read out with an EPI k‐space sampling. The saturation pulse train power (B1_sat) and duration (T_sat) and the excitation pulse FA are varied according to the MRF acquisition schedule. For simplicity, only the saturation power was varied in this study. (B) Schedule of B1_sat powers. (C) Sample CEST‐MRF signals for GM, WM and CSF. (D) DRONE network used in quantification of the CEST‐MRF data. The network is trained with simulated three‐pool data and outputs the water relaxation (T1_w, T2_w), amide (ksw, fs), and semi‐solid (kssw, fss) parameters

FIGURE 2

DRONE reconstruction of six parameters in a digital phantom in comparison to the reference values. Regions associated with the background, skull, and scalp were set to zero. The error, calculated as 100 × |Reference – DRONE|/Reference, is shown for each tissue. Note the effect of the B1 inhomogeneity visible in the error maps

FIGURE 3

(A) NRMSE, on a log scale, of the DRONE reconstructed CEST‐MRF maps in a digital brain phantom using a random schedule for varying levels of added white gaussian noise. Changes in SNR non‐linearly affected the NRMSE of the different parameters illustrating the sensitivity of the sequence and schedule to each tissue parameter. (B) Attribution score calculated using Integrated Gradients, for sample WM and GM fingerprints. Note that the first and last point in the acquisition schedule had the greatest impact on the output

FIGURE 4

(A) Reconstructed tissue parameter maps obtained from a healthy volunteer with the CEST‐MRF method. Note the elevated semi‐solid volume fraction in the WM reflective of the higher myelin content. (B) Reference T1 and T2 maps obtained with the optimized MRF‐EPI sequence

FIGURE 5

Variation in the tissue parameter maps as a function of the B0 inhomogeneity. Shown are the water relaxation parameters: T1_w (A) and T2_w (B); amide parameters: ksw (C) and fs (E ); and semi‐solid parameters: kssw (D) and fss (F). Note the poor correlation between the B0 values and the different parameters illustrative of the robustness of the sequence to B0 variations

FIGURE 6

In vivo GM and WM tissue parameter values for the four CEST‐MRF scans in the healthy volunteer. Scans 1 and 2 were acquired in the first session and scans 3 and 4 in the second session. Blue entries correspond to the left y‐axis and red entries to the right y‐axis with error bars omitted for clarity. (A) T1_w and T2_w. (B) ksw and kssw. (C) fs and fss. The locations of the WM and GM regions used are shown inset in (A). Note the good repeatability between scans. The concordance correlation coefficient for each parameter and tissue type is listed in Table 1

FIGURE 7

Comparison between a measured CEST spectrum and one synthesized from the CEST‐MRF parameters for WM (A) and GM (B). Nuclear Overhauser effects, not included in the CEST‐MRF model, led to the discrepancy between the curves in the negative offsets' region. The measured and synthetic curves were nevertheless highly correlated (r = 0.98) with an RMSE of 0.023 for WM and 0.045 for GM. A comparison between the measured CEST spectrum and one synthesized from the MT and water parameters alone is shown for WM (C) and GM (D). A comparison between a synthetic CEST spectrum and one including only MT and water parameters is shown for in (E) for WM and in (D) for GM

FIGURE 8

(A) In vivo CEST‐MRF maps from a patient with brain metastasis and the corresponding images from a standard clinical protocol for comparison. Green arrows indicate the location of the lesion. The segmented tumor regions are denoted by the colored outlines on the T1_w map and include the edema (black), solid core (blue), necrotic core (red), and contra‐lateral (green) regions. The T1‐pre and T1‐post denote the T1‐weighted acquisition before and after contrast injection whereas Ktrans and Vp refer to the perfusion and plasma volume maps. The marked differences in the tissue map values between the lesion and healthy tissues are notable. (B) Measured and synthetic fingerprints for different tumor regions. The synthesized fingerprints were calculated from the DRONE reconstructed tissue parameters. Note the agreement between the curves. (C) Box and whiskers plots of the reconstructed tissue maps values for the different ROIs. The distribution of the parameter values along with the median and the first and third quartile ranges are shown. (D) Graphical illustration of the statistical significance of the differences between the various tumor ROI pairs. All regions denoted in green were statistically significantly (P = 0.05) as determined by a multi‐comparison analysis of variance test with Tukey honest significant difference

FIGURE 9

Box and whiskers plots for CEST‐MRF‐derived T1, T2, ksw, kssw, fs, and fss as well as conventional sequence values from T1 post Gd contrast, ADC, and FLAIR images. Patient #1: 61‐y‐old male, non‐small cell lung cancer adenocarcinoma, Patient #2: 65‐y‐old female, melanoma, Patient #3: 48‐y‐old female non‐small cell lung cancer adenocarcinoma/small cell, Patient #4: 40‐y‐old male, melanoma (L: left side with two tumors and surrounding region. R: right side with two tumors and surrounding regions)