Parameter Assessment by Retrieval from Signal Encoding (PARSE) is a multi-parameter imaging technique [1]. PARSE simultaneously extracts relaxation rate (R₂*), local frequency offsets (δω), and proton density (M₀) maps from a single FID by utilizing a longer acquisition window to exacerbate R₂* decay and ω shifts in the signal received by the scanner. The speed of PARSE (65ms per image) and its ability to measure phase shifts make it ideal for measuring temporal BOLD fluctuations (used in reactive Oxygen Extraction Fraction (OEF) calculations). Here we develop an improved PARSE reconstruction algorithm that prevents local minima solutions and provides higher quality images. For measurement, we use a rosette trajectory--characterized by fast oscillating frequency (ω₁ = 3874.8 Hz) and a slow rotating frequency (ω₂ = 1610.8 Hz). PARSE provides us with the ability to measure high Signal-to-Noise (SNR) phase accrual because of the frequent re-sampling of the low frequency/center of k-space. An iterative Progressive Length Conjugate Gradient (PLCG) algorithm is used to solve a discretized form of Equation 1. $$ S(t) = \int\int{M_0(x,y)e^{-(R_2^*(x,y)-i\omega(x,y))t}e^{-2\pi i(k_xx+k_yy)}dxdy} $$ Between every iteration of PLCG the result is perturbed and the solution is again allowed to relax using conjugate gradient. At each iteration the least squares estimate is obtained by minimizing the residual between the observed signal and model estimate. The estimate is determined to be adequate when the same minimal square error is reached on 3 separate PLCG iterations. This occurred within (10.4 ± 4.3) iterations.Utilizing the sequence described above, ten hemodynamically compromised subjects (M/F 5/5, <age> = 58.2 ± 9.9 years) and 3 healthy volunteers (M/F, 2/1, <age> = 31.3 ± 2.5) were tested. For each subject we acquired a single slice 2D image (5.0 mm thick, 220 mm x 220 mm FOV, 96x96 matrix, resolution = 2.3 x 2.3 x 5.0 mm3). In one healthy volunteer we acquired five slices separated by 15 mm from the ventricles up. FID data were exported and reconstructed offline. The iterative PLCG reconstruction sum squared signal error was compared against that of the more standard PLCG algorithm. To examine the reconstruction effect on OEF reactivity, estimated local frequency offsets were transposed to OEF using a linear relationship (assuming constant Hematocrit of 18 p.p.m.) [2]. Student’s t-test analysis shows a statistically significant reduction of sum squared error in the iterative PLCG reconstruction vs the more standard PLCG (p< 0.001, Figure 1a) with an average error reduction of 55%. The more accurate fitting results in better anatomical images (shown in Figure 1b). Iterative PLCG yielded improved measured mean OEF in non-affected normal brain parenchyma to 36.87 ± 6.6% and showed affected regions in symptomatic patients reaching 84.05 ± 4.54%--both of which correlate well with literature. Reconstruction has shown to be quite robust even in regions with large susceptibility artifact (Figure 2).We have found that performing iterative PLCG dramatically improves reconstruction quality of an MR-PARSE acquisition. Moreover, iterative PLCG has shown to be capable of reconstruction in regions with large susceptibility artifact (near earholes and sinus, Figure 2). For the purpose of OEF measurements, our primary goal, stable and accurate measurements of frequency shifts are vital. Consistent frequency measurements allow us to remove static offsets caused by air interfaces near the earholes and sinuses leaving dynamic frequency offsets which transpose linearly to OEF. Furthermore, our approach to prevent local minima solutions, through the use of an iterative PLCG, represents a new approach that may improve upon other complex reconstruction methods.