Fixing the MRI R2‐iron calibration in liver

校准 医学 放射科 核医学 数学 统计
作者
Eamon K. Doyle,Nilesh R. Ghugre,Thomas D. Coates,John C. Wood
出处
期刊:American Journal of Hematology [Wiley]
卷期号:95 (5): E120-E122 被引量:9
标识
DOI:10.1002/ajh.25754
摘要

Iron overload is surprisingly common, resulting from genetic abnormalities of iron regulation or as a result of chronic transfusion therapy. The magnetic resonance imaging (MRI) assessment of tissue iron stores has become the standard of care for monitoring iron chelation strategies.1 Note, MRI relaxometry using R2 or R2* are most commonly used. Only one MRI method, based upon a standardized protocol of spin-echo acquisitions and analysis (Ferriscan®, Resonance Health, Western Australia), has achieved regulatory approval in Europe and the United States.2, 3 The Ferriscan® method has been compared against 338 biopsies in two large cohorts, demonstrates good interstudy reproducibility, and has strong quality control practices. However, its cost remains a challenge for many institutions, making iron measurements by R2* acquisitions more financially attractive. Several large studies comparing liver iron concentration (LIC) by R2* and by Ferriscan® R2 have identified substantial bias between R2* and R2 LIC estimates.4-6 We postulated that the original Ferriscan R2 calibration overestimates LIC at high iron concentrations, exaggerating disagreements between the two techniques. We searched the literature for all studies comparing single spin echo R2 acquisitions and liver biopsy results, identifying three studies having 1052, 247, and 2333 liver biopsies, respectively. We used a publicly available program(www.arizona-software.ch/graphclick) to digitally capture the values of R2 for each LIC. We fit the data to the existing FDA-approved calibration2, 7 and compared the residual errors to two other calibration curves. The first was derived using a linear fit in log transformed LIC and R2 coordinates (so-called power-law fit). The second was derived from a spline fit to data generated from previously published8 computer model. This computer model generates "synthetic" R2-iron pairs over the entire physiological range of iron overload, using ideal mathematical approximations to the MRI imaging physics and quantitative statistics of tissue iron deposition.8 It has been used to successfully translate the R2 and R2* liver calibrations to 3T9 with high accuracy. Figure 1A demonstrates a scattergram of all available R2-LIC pairs for liver biopsy data and the FDA approved R2-iron calibration curve. At first blush, the FDA approved calibration appears to represent a good fit to the aggregate data. However, on closer inspection, a preponderance of points lie above the fit line at high LIC. Furthermore, the measurement uncertainty increases as iron burden and liver R2 increases. This is common in biological systems and indicates that calibration error should be calculated as a percentage, rather than an absolute LIC difference. Figure 1B demonstrates the relative difference between the biopsy and FDA-approved LIC plotted against the average of the two measurements. The 95% confidence intervals of the raw Bland Altman relationship are [−69% to 69%]. However, there is a significant downward linear drift (r2 = 0.097, P < .001) with FDA-approved calibration overestimating biopsy by 1.1% per mg/g; the root mean squared of this regression is 30.5%. The drift remained significant even if LIC values greater than 20 mg/g were suppressed. These data suggest that the FDA-approved R2-iron calibration used by Ferriscan® overestimates true liver iron concentration for LIC values exceeding 16.5 mg/g dry weight, with the differences growing geometrically. The calibration error is sufficient to completely explain the differences between LIC by R2* and Ferriscan® R2 described in previously studies.4-6 Importantly, any attempts to "calibrate" R2* or other MRI methods against Ferriscan® need to account for this bias. Several factors contribute to the bias in FDA-approved calibration. The original calibration study probably had insufficient patients (N = 104) to fully characterize a complicated, nonlinear relationship having four degrees of freedom (the power-law calibration has two degrees of freedom). Secondly, the initial patient pool did not have sufficient numbers of patients with severe iron overload. Thirdly, the original calibration was intentionally biased for accurate low iron behavior by including 32 subjects having normal liver iron concentration. It is challenging to find a single curve that perfectly describes very low and very high LIC values. The properties of normal liver structure dominate R2 values at low iron concentration while the impact of siderosomes dominate at high iron concentration.8 The power-law and simulations are optimized for LIC values greater than 5 mg/g while the FDA-calibration was optimized for excellent low iron performance; a piecewise approach may ultimately represent the optimal solution. Lastly, the data heteroscedasticity was not adequately controlled during the fitting and evaluation processes for the FDA-approved calibration, contributing to systematic bias; the power law calibration corrects for heteroscedasticity. Fortunately, the clinical impact of the observed calibration bias is manageable. Patients with LIC below 16.5 mg/g are being accurately risk stratified. Patients with LIC greater than 16.5 mg/g in their liver are universally at high risk and should be treated aggressively, regardless. Serial trends in LIC-R2 values may also lessen the impact of calibration bias.5 Both Ferriscan® and R2* LIC estimates and provide accurate estimates of chelator efficiency on an annual basis.5 Further, both Ferriscan® and liver R2* LIC estimates are more accurate than liver biopsy in tracking changes in total body iron concentration.10 However, hematologists should treat Ferriscan® predicted LIC values of more than 16.5 mg/g with appropriate caution and integrate the predicted LIC values with the patient's entire clinical picture to avoid over-reacting to large changes in predicted LIC. If more accurate LIC predictions are desired in highly loaded subjects, Ferriscan® R2 values can be converted to LIC values from equation (2) using a simple calculator. This work supported by the National Institutes of Health, Diabetes, Digestive and Kidney Diseases (1R01DK097115-01A1). Dr. Wood serves as a consultant to Apopharma, Biomedinformatics, Bluebirdbio, Celgene, Ionis Pharmaceuticals, Imago Biosciences, Silence Therapeutics, and World Care Clinical.
最长约 10秒,即可获得该文献文件

科研通智能强力驱动
Strongly Powered by AbleSci AI
科研通是完全免费的文献互助平台,具备全网最快的应助速度,最高的求助完成率。 对每一个文献求助,科研通都将尽心尽力,给求助人一个满意的交代。
实时播报
刚刚
yyyyy_完成签到 ,获得积分10
1秒前
gytu关注了科研通微信公众号
1秒前
zsy完成签到,获得积分10
1秒前
悦尼酱完成签到,获得积分10
1秒前
俭朴舞仙完成签到 ,获得积分10
1秒前
未了发布了新的文献求助10
1秒前
天真酸奶发布了新的文献求助10
1秒前
香蕉黑米完成签到,获得积分10
2秒前
ace完成签到,获得积分10
2秒前
3秒前
3秒前
青桔子完成签到,获得积分10
3秒前
迷人大凄发布了新的文献求助10
3秒前
Yuuu完成签到,获得积分10
3秒前
迷路鞋子的应助被Maestro_S采纳,获得10
3秒前
科研通AI6.4的应助被喜悦傲霜采纳,获得10
3秒前
3秒前
无敌幸运星完成签到,获得积分10
3秒前
annie完成签到,获得积分10
5秒前
5秒前
5秒前
大个的应助被铁锤牛马版采纳,获得10
5秒前
完美世界的应助被阿飞采纳,获得10
5秒前
搞怪的芷波给搞怪的芷波的求助进行了留言
5秒前
停騮_ 完成签到,获得积分10
6秒前
6秒前
6秒前
个性乐儿完成签到,获得积分10
6秒前
香蕉觅云的应助被luchong采纳,获得10
6秒前
6秒前
HM完成签到,获得积分10
7秒前
中国科研第一烂完成签到,获得积分10
7秒前
8秒前
安静完成签到 ,获得积分10
8秒前
刘超D完成签到 ,获得积分10
8秒前
8秒前
8秒前
充电宝的应助被海伦采纳,获得10
8秒前
Mr完成签到,获得积分10
8秒前
高分求助中
(应助此贴封号)【重要!!请各用户(尤其是新用户)详细阅读】【科研通的精品贴汇总】 10000
拉伸高分子熔体中缠结行为的模拟和小角中子散射研究 1700
The World Atlas of Wine 9th edition 1000
奈特基础组织学:配套组织病理学(第4版)(Netter's Essential Histology With Correlated Histopathology) 800
John McDowells Bildungstheorie 800
Social Psychology in the Real World 600
八旗制度の研究 600
热门求助领域 (近24小时)
化学 材料科学 医学 生物 纳米技术 计算机科学 工程类 有机化学 化学工程 生物化学 物理 复合材料 内科学 催化作用 光电子学 人工智能 细胞生物学 无机化学 物理化学 数学
热门帖子
关注 科研通微信公众号,转发送积分 8180586
求助须知:如何正确求助?哪些是违规求助? 9570581
关于积分的说明 21813457
捐赠科研通 7776777
什么是DOI,文献DOI怎么找? 3433294
关于科研通互助平台的介绍 2557547
邀请新用户注册赠送积分活动 2479253