| Magnetic resonance imaging (MRI) is a medical imaging technique used in radiology to investigate the anatomy and function of the body in both health and disease. MRI scanners use strong magnetic fields and radiowaves to form images of the body. Compared with other modalities of medical imaging, such as conventional X-ray, computer tomography (CT) and ultrasound, MRI has a lot of advantages, such as imaging with an arbitrary plane and with multiple parameters, no ionizing radiation and so on. Moreover, MRI can non-invasively produce images of structure and function of in-vivo tissues in living body, it is thus widely used in hospitals for medical diagnosis, staging of disease and becomes one of the most important methods for the medical diagnosis and scientific research.MRI has emerged as a reliable and non-invasive tool for detecting hepatic iron concentration (HIC) for patients with thalassaemia. Thalassemia is a common hereditary diease which that is characterized by an increase in the gastrointestinal absorption of iron. Repeated blood transfusion therapy has greatly prolonged survival and improved the quality of life for thalassaemia patients in particular those with thalassemia major. However, the transfusion therapy can lead to tissue iron overload which may damage the heart, liver and some endocrine organs over time. Chelation therapy can remove excessive tissue iron from the body and reduce the risk of organ failure in these patients, but may produce toxicity and lead to complications. As HIC is a proven indicator of total body iron store, repeated monitoring of liver iron content is thus vital for assessing treatment effect and minimizing the toxicity associated with iron chelation.There are several methods for the measurement of HIC, such as the plasma ferritin concentration, liver biopsy and consuperconducting quantum interference devices (SQUID). However, the plasma ferritin concentration is influenced by infection, inflammation, liver disease and other factors. Measurement of the iron concentration in a liver-biopsy specimen is the reference method for assessing body iron stores, but it is invasive and potentially unreliable due to sampling errors caused by small biopsy samples. The SQUID device is not in widespread use because of its expensive cost.Magnetic resonance imaging (MRI) has emerged as a reliable and noninvasive tool for detecting iron content. The paramagnetic property of tissue iron causes more rapid signal decay, meaning increased MR relaxation rates Rl, R2and R2*(the reciprocal of Tl, T2and T2*relaxation times, respectively), and to note, good correlation between biopsy-proven HIC and R2or R2*has been observed. Although the R2method for HIC quantification is well established, the R2*method, which utilizes a multi-echo gradient-echo sequence, is generally employed in the clinical area due to its speed, sensitivity, and wide availability. Technically, the R2*method has demonstrated good inter-study and inter-centre reproducibility. Clinically, calibration curves have been derived to convert the measured R2*value to HIC.The ideal relaxation decay curve can be modeled by a monoexponential function. However, it has been observed that the relaxation decay is composed of two components:a fast decay component at early echo time and a slow one at later echo time. This phenomenon can be clearly observed with a high relaxation rate caused by heavy iron overload in the liver. At this scenario, the monoexponential and offset model can not accurately deduce the underlying relaxation process. As is well known that the noise in the MR images approximately follows Rician or non-central Chi distribution because of the root-sum-square reconstruction. A recent study showed that the offset model severally overestimated and underestimated R2*at poor signal-to-noise ratios (SNRs), which frequently occurs in the image of liver with severe HIC; fitting the noisy signal to its first moment or fitting the squared noisy signal to the second moment in the presence of non-central Chi noise (first moment noise-corrected model, M1NCM; second moment noise-corrected model, M2NCM) could produce accurate R2*measurements even at high noise levels. Compared to the M2NCM, the M1NCM produced lower variance of R2*.For the liver R2*measurement, a representative R2*value of one or more regions-of-interest (ROI) is typically calculated and converted to average iron content by using a calibration curve in the clinical practice. To exclude blood vessels and other focal liver lesions, multiple ROIs (mROI) need to be carefully delineated in the homogenous parenchyma area by checking image of each echo time (TE). After delineating mROI, the signal intensities of all pixels in the mROI for each TE image are usually averaged and then fitted with an appropriate decay curve model to generate the representative R2*value (average-then-fit, ATF). Alternatively, the mean (PWFmea) or median (PWFmed) R2*value of the mROI in a R2*map acquired by pixel-wise fitting (PWF) is reported. Although the mROI approach has shown good correlation with biopsy specimen HIC, operator-dependent placement of the ROIs and the potential uneven distribution of HIC may lead to measurement errors.In recent years, the "whole liver" ROI R2*measurement, in which the ROI outlines the whole liver except large vessels in a single slice, has been shown to be more reproducible than the small ROI method. However, manually extracting the liver parenchyma from a whole liver ROI is challenging due to the irregular distribution of blood vessels. Thus, several computer-aided methods have be proposed for accurate and efficient liver parenchyma extraction by T2*thresholding method and adaptive fuzzy-clustering algorithm. However, the T2*thresholding was dependent on the subjectively determined threshold; and PWF in the liver parenchyma region may be degraded by low signal-to-noise ratio (SNR) and produces T2*estimates with a large variance which may challenge the selection of an optimal T2*threshold. In addition, the pixels near the margin between the parenchyma and non-parenchyma tissues are usually affected by partial volume effect (PVE), which leads to intermediate T2*values between typical T2*values of parenchyma and non-parenchyma tissues and further increases difficulty in choosing a T2*threshold. The segmentation by the adaptive fuzzy-clustering algorithm is based on only one single TE image and may be unable to exclude the focal liver lesions as hemangiomas which are frequent and sometimes visualized only on images of later echoesIn summary, the R2*mersurement of semi-automatic parenchyma extraction (SAPE) method is proposed. The proposed SAPE method, including three main elements:the NLM filter, the FCM classification and the MP operation, can be implemented in five steps:1) Filter the images by using the NLM algorithm to reduce noise while preserving edges between the liver parenchyma and non-parenchyma.2) Draw a ROI to outline the whole liver in a single slice image; and then calculate the R2*map in the ROI by using the PWF approach.3) Perform the FCM algorithm to classify the R2*values within the ROI into two groups (the parenchyma and non-parenchyma).4) Implement the MP operation on the extracted parenchyma to exclude the pixels near the edges between the parenchyma and non-parenchyma which are likely affected by PVE.5) Given the extracted parenchyma, the ATF approach with the the first moment noise-corrected curve fitting model was adopted on the filtered images to acquire the representative R2*value. To illustrate the indispensability of the NLM prefiltering and the MP post-processing in the proposed method, the proposed SAPE method was compared with the FCM only, FCM+MP, and NLM+FCM approaches in the simulation study.The R2*measurement accuracy of the SAPE method was evaluated through simulation. Numerical simulations were performed to evaluate the accuracy of R2*measurement with the advantage of known true R2*values. In this study, a mask indicating the structure of the liver parenchyma and vessels was derived from a liver image of a single patient which was manually segmented by an experienced observer, and utilized to synthesize a reference R2*map with typical R2*values corresponding to the parenchyma and vessels. The R2*values of parenchyma ranged from100s-1to1000s-1with a100s-1increment, corresponding to HIC from normal to high levels. Given the R2*reference map, the ideal sequential images were synthesized by sampling the ideal free induction signals. Finally, the partial volume effect, intensity inhomogeneity and noise were sequentially added to the ideal images.In this study a total of108transfusion-dependent patients (56males and52females, mean age23±10years), were scanned on a Siemens Sonata1.5T scanner (Siemens Medical Solutions, Erlangen, Germany), and with ethics board approval and written informed consent obtained from all subjects. A single transaxial slice through the centre of the liver was scanned using a multi-echo gradient-echo sequence with the following acquisition protocols:flip angle of20°, repetition time of200ms, TEs of12(0.93,2.27,3.61,4.95,6.29,7.63,8.97,10.4,11.8,13.2,14.6,16ms), slice thickness of10mm, bandwidth per pixel of1955Hz, matrix of64×128, number of averages of1. In-plane resolution ranged from2.2x2.2to3.1x3.1mm2according to the subjects’size. Fat saturation was applied and the multiple TE images were acquired within a single breath-hold of approximate13seconds. The intra-and inter-observer reproducibility of SAPE, mROI and T2*thresholding were assessed from the in vivo data using the Bland-Altman analysis and coefficient of variation (CoV).To our knowledge, there were three approaches (ATF, PWFmea, PWFmed) for the R2*measurement of ROI. The performance of the PWFmed, PWFmea and ATF approaches implemented with the recently-proposed M1NCM model has not yet been compared. Since the PWFmed, PWFmea and ATF approaches may produce potential discrepancy in R2*measurements, which may affect the clinical therapeutic decision-making, this study aims to compare and evaluate the performance of the three approaches implemented with the M1NCM model and determine which of them is optimal for liver R2*measurement by simulations (with known R2*values) and patient studies.The mean absoulte R2*evaluation error percentage of the SAPE method compared to the true R2*was0.23%(ranged0.01%～1.09%). The in-vivo study shows that there was a good correlation between the liver R2*values measured by the SAPE and mROI or T2*thresholding methods with CoV of5.25%and2.75%. The CoV of intra-and inter-observer variability for the SAPE method was0.83%and1.39%, compared with3.63%and6.28%for the mROI method,1.62%and1.39%for the T2*thresholding method. In conclusion, a comprehensive semiautomatic parenchyma extraction method was proposed and effectively applied to liver R2*measurement. The proposed SAPE method may provide a more accurate and reproducible tool for the hepatic iron concentration quantification by significantly reducing the operator dependence, which may lead to more accurate tissue characterisation and increased diagnostic confidence.For the M1NCM curve-fitting model, both the ATF and PWFmed approaches can produce more accurate R2*measurements for HIC quantification, and are highly preferable to the PWFmea approach, which tends to overestimate R2*measurements, particularly under severe HIC. Thus, both the ATF and PWFmed approaches can be used with the M1NCM model to report a representative R2*of the ROI in the clinical practice. Compared with ATF, the PWFmed method has the advantage of providing the spatial distribution of R2*but requires numerous curve-fitting operations to generate an R2*map, thus takes considerately longer time. Future research is needed to establish the calibration curve to convert the R2*values measured by the ATF and PWFmed approaches with the M1NCM model into the liver iron content. |
PDF Full Text Request