Imaging Core Lab Comparison of Five Clinical Liver Iron MRI Methods and Biopsy for Hepatic Iron Concentration (HIC) in Sickle Cell Disease (SCD) Patients.

Introduction: HIC is a surrogate for total body iron in pts with transfusional iron overload. MRI techniques based on signal intensity ratios (SIR) or relaxation rates (R2 and R2*) are increasingly accepted as surrogates to biopsy in determining HIC. MRI techniques have been validated using data acquired on different MRI platforms, but image processing has been restricted to labs where the techniques were developed. We tested the ability of an imaging core lab to process data from 4 different MRI techniques:

1. Gradient echo (GE) SIR (Gandon et al);

2. Spin echo (SE) SIR (Jensen et al);

3. R2 (Wood et al);

4. R2* (Wood et al).

R2 St. Pierre processing is proprietary and was not implemented by the core lab. The goal was to collect pilot data to test the feasibility of independent core lab assessment as well as assess inter-technique agreement and interobserver variability.

Methods: Seven CICL670A0109 MRI/liver biopsy substudy pts (all at CHLA; one withdrew after baseline [BL] exam) participated in an open-label comparison of deferasirox and DFO in SCD pts. Liver biopsy (analyzed centrally at Ctre Hosp Univ, Rennes, France) was performed at BSL and 1 yr, and MRIs at BL, 6 mo and 1 yr on a 1.5 T GE CVi scanner, system 9.1. Exam consisted of localization, a total of 15 breathhold acquisitions for GE-SIR, R2* and R2 (Wood), and 30 min of free-breathing SE acquisitions for SE-SIR and R2 (St Pierre). A MnCl₂ phantom (13 vials, 0–24 mM) was scanned 5x over 14 mo. De-identified DICOM images for SIRs, R2* and R2 Wood were analyzed at an imaging core lab. R2 and R2* maps were reconstructed by pixelwise fitting to a monoexponential plus an offset using custom (IDL) software. ROI selections were validated with the investigators responsible for developing the techniques (YG, PJ, JW).

Results: BL biopsy HIC was 18.8±5.7 [11.7–28.3]. High HIC precluded successful HIC calculation by GE-SIR in 17/19 examinations. In contrast, SE-SIR was measurable in all studies but demonstrated a large bias compared with biopsy (Table). All three relaxation rate techniques yielded unbiased HIC assessment with acceptable agreement. Unlike previous reports, the MRI techniques agreed better among themselves than with biopsy (R=0.75–0.93), suggesting that, even with standardized collection, biopsy variability represented a major source of error in this study. Interobserver variability had means=−1.5–2.6% and SD=1.9–2.4 %. Errors introduced by different fitting algorithms had mean values of 0.5%–1.4% and SD of 1.3–3.1%. MnCl₂ phantom MR values were stable over study duration (COV=3.5% [1.2–8.0]).

Conclusion: With higher HIC, R2 and R2* metrics are more generalizable than SIR metrics. Computational and observer-based errors introduced by differences in image post-processing are small compared with the errors introduced by biopsy sampling errors and intersubject variability in MRI-iron calibration.