Analysis of signal dynamics in oxygen-enhanced MRI

Introduction: Oxygen-enhanced MRI (O2-MRI) has been successfully applied for the visualization and (semi-)quantitative measurement of pulmonary ventilation and lung function [1–6]. The inhalation of pure oxygen (O2) decreases the longitudinal relaxation time, T1, of blood, which can be detected by T1-weighted MRI. Frequently, a block paradigm is used for O2-MRI consisting of a series of T1-weighted scans acquired during alternating blocks with inhalation of room air and O2. This block design results in a signal-time course for each pixel that contains information about lung function as well as respiration and circulation parameters. The purpose of the present study was to analyze the properties of this signal-time course of O2-MRI in more detail by evaluating several model functions with different parameters. Methods: We studied 11 healthy volunteers using a T1-weighting multi-slice inversion-recovery halfFourier-acquisition single-shot turbo-spin-echo (HASTE) sequence (TI=1300 ms, TE=11 ms, 4 slices, slice thickness 8 mm, slice distance 16 mm, matrix 128×128, FOV 400×400 mm2, GRAPPA acceleration factor 2) implemented on a 1.5-T whole-body scanner (Magnetom Sonata, Siemens Healthcare, Erlangen, Germany) with an 8-channel phased-array thorax coil system. Each examination consisted of a series of 60 acquisitions (20×air, 20×O2, 20×air) with ECG and respiratory triggering for acquisition in end-expiration. Lung tissue and the spleen were segmented manually in all 11 data sets (44 slices). We normalized all data pixelwise to their mean value over the 20 initial baseline (air) acquisitions to obtain relative enhancement data. We compared six different model functions (a rectangular, a linear, and four exponential functions, cf. Fig. 1) for the evaluation of the signal-time course in all data sets. The model functions were fitted to the data from acquisitions 21...60, i.e. to the O2 wash-in and wash-out periods; non-linear fitting was performed with a Levenberg-Marquardt implementation. The optimal fit function was determined by the corrected Akaike information criterion (AIC) [7], which is a measure based on the sum-of-squares difference (SS) between data and fitted function, the number, N, of data points, and the number, K, of free parameters: AIC = N ln(SS/N) + 2(K+1) + 2(K+1)(K+2)/(N–K–2). The optimal fit function (i.e. the one with the minimal AIC) was then used to calculate maps and median values of all parameters. Results: The model function with the minimal AIC (cf. Fig. 2) was the 4-parameter exponential model (4P_exp). The median values of the fit parameters in all 11 volunteers are summarized in Table 1. There is no position dependence of the parameters within the lung in anterior-posterior direction. Noteworthy are the substantial differences of the time delay, Δt, in the lung and the spleen as well as the prolonged wash-in time constant in the spleen. Typical parameter maps are shown in Fig. 3. Table 1: Parameters of oxygen dynamics in healthy volunteers (median values, 16th...84th percentile) c (%) Δt (s) τ1 (s) τ2 (s) Lung slice 1 (anterior) 16.3 (7.6...30.4) 6.3 (0.0...25.2) 33.8 (2.1... 96.0) 26.4 (0.5...92.5) Lung slice 2 17.0 (9.2...29.9) 5.0 (0.0...25.3) 31.5 (3.4... 84.0) 26.7 (5.8...71.3) Lung slice 3 15.9 (9.4...25.9) 6.5 (0.0...18.2) 28.8 (8.5... 71.7) 25.4 (3.8...61.2) Lung slice 4 (posterior) 16.3 (10.7...27.8) 0.9 (0.0...13.6) 27.3 (10.7... 77.3) 23.3 (8.3...53.8)