Received date: November 22, 2011; Accepted date: January 24, 2012; Published date: January 27, 2012
Citation: Li W, Du H, Prasad PV (2012) Would Sub-regional Analysis Improve Sensitivity in Knee dGEMRIC? J Mol Imag Dynamic 2:105. doi:10.4172/2155-9937.1000105
Copyright: © 2012 Li W, et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Visit for more related articles at Journal of Molecular Imaging & Dynamics
Purpose: To evaluate if sub-regional analysis could offer higher sensitivity than full-thickness analysis when quantifying dGEMRIC data of the knee. Materials and methods: A dGEMRIC data set of medial femoral cartilage of the knee, including thirteen patients with osteoarthritis (diseased) and fourteen asymptomatic subjects (control), was reanalyzed with sub-regional analysis, i.e. using two ROIs (regions of interest) representing superficial region and deep region of the cartilage respectively, in addition to full-thickness ROI. Comparisons were made in T 1pre (spin-lattice relaxation time without contrast administration), T 1Gd (after Gd-DTPA 2- administration), and ΔR 1 (the change of relaxation rate) obtained with the three analysis methods. Results: Differences between superficial region and deep region in T 1pre , and T 1Gd were clearly observed. Superficial region analysis always provided shortest T 1Gd and highest ΔR 1 in both the diseased and the control, but the diseased-control difference in T 1Gd and ΔR 1 were quite close with the three analysis methods. Even though superficial region analysis showed slightly higher values of AUC (areas under the curves) of Receiver Operating Characteristic compared to full-thickness analysis, no significant difference in sensitivity was observed for knee dGEMRIC, with either T 1Gd (p=0.802) or ΔR 1 (p=0.328). Conclusion: For analyzing dGEMRIC of the knee with reduced cartilage thickness due to OA, there is no significant difference in sensitivity between sub-regional analysis and full-thickness analysis.
dGEMRIC; Contrast agent; Sub-regional analysis; Knee; Osteoarthritis
Spin-lattice relaxation time (T1) of articular cartilage after contrast administration of Gd-DTPA2- (T1Gd) has been commonly used as a dGEMRIC index to determine relative glycosaminoglycan (GAG) level within articular cartilage [1-3]. It has been argued that tracking ΔR1, i.e. the change in relaxation rate (R1 = 1/T1) before (R1pre) and after Gd-DTPA2- administration (R1post), is necessary . A previous report showed that T1Gd and ΔR1 were similarly effective in differentiating knees with osteoarthritis (OA) from the healthy . To-date, quantification of T1Gd and ΔR1 has been based on the analysis of regions of interest (ROIs) covering full thickness of cartilage.
It has been reported that the GAG concentration profile in canine articular cartilage was tissue depth dependent . Degeneration of cartilage in aging and osteoarthritis generally progresses from the surface of the cartilage . A recent report on asymptomatic subjects  suggested that when performing an analysis of the knee based on the depth of the cartilage, the superficial layer of articular cartilage showed longer T1pre, shorter T1Gd, and hence larger ΔR1 compared to the deep layer. These reports then put forward a practically interesting question: would dGEMRIC analysis be more sensitive if sub-regional ROI of cartilage, instead of full thickness of ROI, was used to quantify the biomarkers for assessing health status of articular cartilage?
In this study, we reanalyzed a dGEMRIC data set  of the knee with sub-regional analysis, i.e. using two ROIs representing superficial region and deep region of the cartilage respectively, in addition to fullthickness ROI. The effectiveness across the three analysis methods: superficial region analysis, full-thickness analysis, and deep region analysis, was compared. Since the ROIs we used did not exactly represent any single anatomical layer, we used the word “region” instead of “layer” in this study to avoid misunderstanding.
We performed a retrospective review of knee dGEMRIC datasets of eighteen patients with OA and fourteen asymptomatic subjects. One knee was imaged for each case. Of the eighteen patients, five patients were excluded from the study due to limited cartilage available in which no more than one ROI could be accommodated. The rest thirteen patients (diseased group, 5 males, average age of 60.8 ± 13.0 years) and the fourteen asymptomatic subjects (control group, 5 males, average age of 29.2 ± 7.4 years) were reanalyzed using ROIs covering superficial region and deep region of cartilage respectively, in addition to full-thickness coverage. OA was diagnosed by clinicians based on clinical symptoms only (n=4), or symptoms along with X-rays (n=8), or symptoms along with MR imaging (n=1). For asymptomatic subjects, there was no symptom with the knee and no history of documented knee disease or injury. The study was conducted in compliance with the regulations of the institutional review board and informed consent was obtained from each subject.
All subjects were imaged before and 90 min after double-dose Gd- DTPA2- (0.2 mmol/kg) administration on a 1.5 Tesla MR scanner (GE Healthcare, Waukesha, WI) with a transmit/receive extremity coil. A two-dimensional inversion recovery fast spin-echo (2D IR-FSE) sequence was used for all exams, except for nine pre-contrast exams (control = 5; diseased = 4) in which a three dimensional Look-Locker (3D LL) sequence was applied. The 3D LL technique was shown to provide accurate T1 estimates compared to 2D IR-FSE method [9,10]. The 2D IR-FSE scans in single slice were positioned sagittally going through the medial femoral condyle. The nine pre-contrast 3D LL scans were positioned sagittally covering whole joint, but only the slice through femoral medial condyle, which was matched with the position of post-contrast 2D IR-FSE image, was used for analysis. Imaging parameters relevant to spatial resolution were: slice thickness = 3 mm, field of view = 160 mm, Matrix was 3842 (in plane resolution = 417μm, interpolated to 5122) for post-contrast imaging, and was 2562 (in-plane resolution =625μm, interpolated to 5122) for pre-contrast imaging. Other imaging parameters can be referred to our previous study .
T1 mapping and data analysis
T1 maps of the medial femoral cartilage were generated using a custom software analysis package (MRIMapper, copyright Beth Israel Deaconess Medical Center and Massachusetts Institute of Technology, 2006) written in MATLAB (The Mathworks; Natick, MA). ROIs were segmented at superficial region, deep region, and full thickness of the medial femoral cartilage respectively to obtain T1 values of pre- (T1pre) and post-Gd-DTPA2- administration (T1Gd). For better representing the health status of the cartilage, in sub-regional analyses we applied relatively thicker (particular thickness in each case depended on the cartilage available) and larger (covering the weight-bearing area as much as possible) compared to previous report . For control group, the sizes of ROI (in number of pixels) were 129 ± 31, 162 ± 37, 538 ± 113 in pre-contrast images, and 135 ± 37, 154 ± 46, 570 ± 164 in post contrast images, in the order of superficial region, deep region, and full thickness respectively. For diseased group, the corresponding numbers were 85 ± 41, 119 ± 45, 385 ± 170 in pre-contrast images, and 88 ± 34, 121 ± 36, 341 ± 162 in post contrast images. The attention was paid to avoid partial volume effect with adjacent tissues and to make superficial and deep regions parallel to each other as much as possible (Figure 1). The average T1 value of all pixels within the ROI was used. The change in relaxation rate (ΔR1) between post- and pre-contrast administration was calculated as 1/T1Gd - 1/T1pre.
The two tailed t-test was used for statistical analysis. A mixedeffects regression analysis was performed to assess whether the sensitivity varied across the three analyses methods. Unstructured variance-covariance was used to adjust within-subject correlation. A logistic regression was also applied to generate areas under the curves (AUC) of Receiver Operating Characteristic, with corresponding optimal threshold values (best-cut) for separation of the diseased and the control .
Table 1 lists the values of T1pre, T1Gd, and ΔR1 of the medial femoral cartilage obtained with the three analysis methods for each subject. The difference between superficial region and deep region in T1pre, and T1Gd could usually be visualized on T1 maps (Figure 2). Compared to control group, diseased group always showed statistically significant shorter T1Gd and higher ΔR1, irrespective of which analysis method was used.
|T1 pre (ms)||T1Gd (ms)||ΔR1 (s-1)|
|Average SD (in%)||993 156 (16)||1050 225 (21)||933 139 (15)||391 81 (21)||335 85 (25)||441 82 (19)||1.65 0.65 (39)||2.18 0.78 (36)||1.26 0.59 (47)|
|Average SD (in%)||905 52(6)||985 82(8)||814 57(7)||517 69(13)||458 69(15)||570 85(15)||0.86 0.30(35)||1.21 0.35(29)||0.57 0.35(62)|
|P - Diseased vs Control||>0.05||>0.05||<0.05||<0.001||<0.001||<0.001||<0.001||<0.001||<0.01|
* With t-test; Full: full-thickness analysis; Super: superficial region analysis; Deep: deep region analysis
Table 1: T1pre, T1Gd and ΔR1 of femoral cartilage measured with the three analysis methods.
Figure 2: Demonstration of T1 differences between deep and superficial regions before and after Gd-DTPA2- administration in a healthy subject. Note that in the overlaid T1 map before contrast administration (Pre-Gad), T1-superficial is higher than T1-deep; whereas in T1 map after contrast administration (Post- Gad), T1-superficial is lower than T1-deep
Table 2 and Figure 3 summarized the comparison results across the three analysis methods. Superficial region analysis always provided shortest T1Gd and highest ΔR1 in both the diseased and the control, which was followed by full-thickness analysis and then deep region analysis (Figure 3a, b), but the differences between the diseased and the control (diseased-control difference) were quite close in the three analysis methods. With T1Gd, average values of diseased-control difference (T1Gd of the control - T1Gd of the diseased) were 123, 126, and 129 ms for superficial region, full-thickness, and deep region analyses respectively. The three variation lines were almost parallel to each other (Table 2 and Figure 3c). With ΔR1, the corresponding diseasedcontrol differences (ΔR1of the diseased - ΔR1 of the control) were 0.97, 0.79, and 0.69s-1, and variation line of superficial analysis was slightly steeper compared to other two lines (Table 2 and Figure 3d). Standard deviations with ΔR1 were much larger compared to T1Gd (Table 1 and Figure 3a, b). In both T1Gd and ΔR1, superficial region analysis showed slightly higher AUC, followed by full-thickness analysis and then deep region analysis. But all three AUCs are larger than 0.8 or 0.9, indicating excellent or distinguished discrimination . The optimal threshold values are different with the three methods. Mixed-effects regression analysis demonstrated that there were no significant differences across the three analysis methods in differentiation of the diseased from the control (Table 2).
|with T1Gd (ms)||with ΔR1 (s-1)|
Full: full-thickness analysis; Super: superficial region analysis; Deep: deep region analysis;
D-C Difference: Difference between the diseased and the control; D-C best-cut: the best threshold for separation of the diseased and the control; AUC: areas under ROC curve;
* with mixed-effects regression analysis
Table 2: The comparison of the three analysis methods.
Figure 3: (a) and (b) show the average T1Gd and ΔR1 of the two groups obtained with the three analysis methods: superficial region analysis provides significantly shortest T1Gd and highest ΔR1 in both the diseased and the control, followed by full-thickness analysis and deep region analysis; much larger standard deviations were observed with ΔR1 compared with T1Gd. (c) and (d) demonstrate the variation lines (line connecting the two average values of diseased and control groups) of the three analysis methods. With T1Gd the three lines are almost parallel to each other, indicating the three analysis methods are almost equally effective. With ΔR1, the line of superficial analysis is just slightly steeper compared to other two lines, but no significant difference was observed because of higher standard deviation.
The anatomical feature of the depth wise variation of articular cartilage is well known. Native articular cartilage is composed of an extensive extracellular matrix synthesized by chondrocytes. It contains different zones with respect to depth from the articular surface and has a regional organization around the chondrocytes. Its composition varies regionally and zonally in its collagen and proteoglycan contents, and those of other matrix molecules . GAG concentration profile has been reported to vary approximately linearly as a function of the tissue depth in canine articular cartilage . A number of studies have investigated the spatial distribution of T2 and T1ρ in articular cartilage [13,14]. A study on asymptomatic subjects has demonstrated the difference between superficial layer and deep layer in T1pre, T1Gd, and ΔR1 of articular cartilage . But, to our best knowledge, using subregional analysis for dGEMRIC for OA and asymptomatic subjects has not been published.
In this study, the variations between superficial region and deep region in T1pre and T1Gd could usually be visualized on the T1 maps, which was consistent with previous report . The variation in T1pre may reflect the differences in molecular structure and composition between regions. The T1Gd disparity between regions may indicate additional differences in GAG level and other factors, such as transport of contrast to respective regions [8,15].
We had expected that using superficial region analysis, instead of full-thickness analysis, might have advantage for dGEMRIC to get better sensitivity. Because of the depth wise variation of articular cartilage and that the degeneration of cartilage in aging and osteoarthritis generally starts from the surface of the cartilage , superficial region analysis could show shorter T1Gd in diseased cartilage compared to full thickness analysis, and hence lead to larger diseased-control differences in both T1Gd and ΔR1. However, our data did not show obvious increase in diseased-control difference with sub-regional analyses in either T1Gd or ΔR1 compared to full-thickness analysis. Even though superficial region analysis showed slightly higher AUC values compared to full-thickness analysis, no significant difference in sensitivity across the three analysis methods was observed with either T1Gd or ΔR1, as shown in Table 2.
It is possible that the negative result observed in this study may be related to the reduced cartilage thickness in subjects with OA. A limitation of the study is lack of staging information on OA patients. However, our recent experience suggests that even OA subjects with grade of KL-2 could exhibit full thickness cartilage loss . It is also interesting that the recent report in healthy subjects  shows the medial cartilage thickness to be variable from 1-4 pixel wide. The pathologic changes in these cases could exist not only in superficial region of the cartilage, but also in deeper or even the full thickness of the cartilage. Larger pathologic area involved would lead to less difference between superficial region and deep region. As a result, the diseased-control differences of T1Gd with the three analysis methods were almost identical (123 ms, 126 ms, and 129 ms for superficial region, full-thickness, and deep region analyses respectively). Although the diseased-control difference of ΔR1, which was influenced by both T1Gd and T1pre, was slightly higher with superficial region analysis (0.97s-1) compared to full thickness analysis, it did not reach statistical significance probably due to the large variance associated. This result suggests that for subjects with reduced cartilage thickness, sub-regional analysis has comparable sensitivity in either T1Gd or ΔR1 compared to full-thickness analysis. It should be noted that the best-cuts (thresholds) were quite different when using different analysis methods, as shown in Table 2.
The ROIs’ size, range and location and in-plane resolution could also affect the result of dGEMRIC. The result presented here was based on our relatively thicker and wider ROIs, under the in-plane resolution similar to the in-plane resolution commonly suggested for dGEMRIC . In the previous study, which was to investigate the transport of Gd-DTPA2- in different layers , the authors applied one-pixel thickness ROIs and confined the ROIs to a specific area. In this study we used relatively thicker and wider ROIs to better reflect the health status of cartilage as a whole.
The major limitations of this study are that the number of subjects was small and lack of early stage of OA included. A study with larger subject number including enough patients with early stage of OA is warranted, which may show different trends compared to the present study.
In summary, the depth wise variations in T1pre and T1Gd were clearly observed in both diseased and control group. For analyzing dGEMRIC of knee articular cartilage in OA subjects with reduced cartilage thickness, there was no significant difference in sensitivity between sub-regional analysis and full-thickness analysis.
The authors thank Ms. Sally Gartman for editorial help with the manuscript.