To design and develop a new high sensitivity 3D gagCEST technique to quantify glycosaminoglycan present in human knee cartilage in a practically achievable scan time.All the human scans were performed under an approved Institutional Review Board protocol of the University of Pennsylvania. MRI scans were performed on 2 healthy male subjects in the age range 20 – 35 years and one symptomatic male subject aged 65 years old, with knee pain at Siemens 7T whole body MRI scanner (Siemens Medical Solutions, Malvern, PA) using a 28-channel receive array knee coil (Quality Electrodynamics, Mayfield Village, OH). A new 3D ‘burst mode’ sequence was developed as shown in Fig 1. In the burst mode magnetization prepared 3D sequence, the repetition time between magnetization preparation pulses is set to be long (> 3 x T₁). This ensures that the bulk water signal has fully recovered from the previous preparation pulse. Now, to optimize scan time, we used very long low flip angle segmented gradient echo (GRE) readout with elliptically centered ordering for slice encoding (k_z) and phase encoding (k_y). For CEST, a frequency-selective saturation pulse train of 5 Hanning windowed pulses with the duration of 99.8 ms, with a 0.2 ms gap, with a B_1rms of 2.2 µT was used [1]. For B₀ mapping with Water Saturation Shift Referencing (WASSR) [2], a frequency-selective saturation pulse train of 2 Hanning windowed saturation pulses with a B_1rms of 0.3 µT was used. For B₁ map generation, hard pulses with two predefined flip angles are used for magnetization preparation. A new concentration independent B₁ calibration method was developed. Spatial encoding parameters were slice thickness = 3 mm, number of slices = 16, flip angle = 5°, TR/TE = 7.8/3.6 ms, FOV = 140 mm with 0.6 mm² in-plane resolution for both axial and coronal orientations. The total scan time for acquisition in both the axial and coronal orientations was 50 min, which includes positioning of the knee as well as the shimming for B₀ inhomogeneity. In all the datasets acquired, with both axial and coronal orientations, SNR of cartilages in the images with no saturation magnetization was ~95. For the ROI’s chosen from the healthy regions, CEST asymmetry at 1 ppm from steady-state method [3] is ~3% while that for the burst saturation method is ~7%. Representative slice of gagCEST maps with M0 (water magnetization image with no preparation pulse) and Mz- (water magnetization image with preparation pulse corresponding to -1 p.p.m.) normalizations of patellar and femoral-tibial cartilages from a young healthy volunteer and an elderly subject with reported knee pain are shown in Figs. 2A and 2B. The elderly subject’s patellar cartilage was found significantly thinner than the young subject’s cartilage while this thinning was not observed in the femoral and tibial cartilages. The mean CEST contrast values from the entire cartilage region for both normalizations are shown in Figs. 2C and 2D. Mz- normalization showed a better dynamic range (~9% to ~16%) than the M0 normalization (3.7% to 4.3%). Also, M_z- normalization showed better discrimination between the patellar cartilage of the healthy young subject and the elderly subject (mean gagCEST values, 12.75 ± 4.7% and 9.48 ± 3.7%). The final gagCEST images calculated after motion correction, B₀ and B₁ correction and Mz- normalization are shown in Fig. 3. One can clearly see layer-wise differences in gagCEST maps.The new burst mode magnetization preparation 3D gagCEST method along with separate B₀ and B₁ inhomogeneity estimation and correction, yielded reliable and reproducible high-quality gagCEST maps with clear visualization of knee cartilage layers. The longer repetition times between saturation pulses and elliptical-centric ordered segmented gradient echo (GRE) readout seems to maintain high sensitivity, while in the steady-state method the CEST sensitivity was lower due to the use of very short repetition times between saturation pulses and the addition of a water saturation pulse to maintain constant initial magnetization.