Comparison of MRI T1, T2, and T2* mapping with histology for assessment of intervertebral disc degeneration in an ovine model

An easy, reliable, and time-efficient standardized approach for assessing lumbar intervertebral disc (IVD) degeneration with relaxation times measurements in pre-clinical and clinical studies is lacking. This prospective study aims to determine the most appropriate method for lumbar IVD degeneration (IDD) assessment in sheep by comparing three quantitative MRI sequences (variable-flip-angle T1 mapping, and multi-echo T2 and T2* mapping), correlating them with Pfirrmann grading and histology. Strong intra- and interrater agreements were found for Nucleus pulposus (NP) regions-of-interest (ROI). T1, T2, and T2* mapping correlated with Pfirrmann grading and histological scoring (p < 0.05) except for the most ventral rectangular ROI on T2 maps. Correlations were excellent for all of the T1 ROIs and the T2* NP ROIs. Highly significant differences in T1 values were found between all Pfirrmann grades except between grades I/II and between grades III/IV. Significant differences were identified in the T2 and the T2* values between all grades except between grades I/III. T1, T2, and T2* relaxation times measurements of the NP are an accurate and time-efficient tool to assess lumbar IDD in sheep. Variable-flip-angle T1 mapping may be further considered as a valuable method to investigate IDD and to assess the efficacy of regenerative treatments in longitudinal studies.

. The times of acquisition were 164 s for the T1-weighted fast spin-echo sequence, 231 s for the T2-weighted fast spin-echo sequence, 88 s for the T1 mapping sequence, 306 s for the T2 mapping sequence, and 101 s for the T2* mapping sequence. The nine elements of the spine coil were used to obtain the sequences. An example of the full FOV images for T1, T2, and T2* mapping is provided as Supplementary Data ( Figure S1 in the Appendix). To prevent diurnal variations in the relaxation times, as previously described 28 , the images were systematically acquired in the morning. Image analysis. The image data were analyzed with Horos software 3.0® (Horos Project, Geneva, Switzerland). Radiographs were useful to determine the number of lumbar IVDs in each sheep, as they vary between 5 and 6 in this species 29 , and to verify the absence of major vertebral disease (e.g., malformations, old fracture, or tumors).
Each IVD was assigned a degenerative score on T2-weighted mid-sagittal images using a 5-point grading system as described by Pfirrmann et al. 9 To delineate the various parts of the IVD, ROIs were drawn following two previously described drawing methods: the 3-ROIs method and the 5-ROIs method (Fig. 1) 27 . The size of each ROI varied according to the size of the IVD. In the 3-ROI method, on the T2-weighted images, an oval ROI was manually drawn around the NP (ROI B) and two other oval ROIs were drawn from either side representing the ventral AF (ROI A) and the dorsal AF (ROI C). Particular attention was paid to not include parts of the vertebral endplates in the ROI. The contrast of the images was changed manually in order to distinguish better the NP. In the 5-ROI method, a rectangular ROI was drawn on the T2-weighted images with these limits: between the cranial and the caudal endplates and between the dorsal and the ventral borders of the IVD. This rectangular region was subdivided into 5 equal rectangular ROIs with ROI 1 corresponding to the most ventral ROI and ROI 5 corresponding to the most dorsal ROI. For each IVD, those eight ROIs were copied on the T2 weighted images and pasted on the corresponding T1, T2 and T2* mapping images. The position of each square in the 5-ROI method was manually slightly readjusted in the craniocaudal axis in order not to include vertebral endplates in the ROI section. The mean T1, T2, and T2* relaxation time values of each ROI were collected. Each value was determined by the mapping method (T1, T2, or T2*) and the corresponding letter (3-ROIs method) or number (5-ROIs method). For example, the value of the NP collected by the 3-ROIs method on the T1 mapping images was designated  www.nature.com/scientificreports/ "T1ROIB". The data of the time dedicated to each drawing method was collected by one observer on one IVD of each animal. The time dedicated was measured from the opening of the qualitative MRI image on which the ROIs were initially drawn to the end of the data collection on the three quantitative mapping sequences. The mean time of the ten measures was kept to compare the two drawing methods.
Histological analysis. The animals were euthanized after the MR exam by intravenous injection of 140 mg/ kg pentobarbital. Fifty lumbar discs were harvested and fixed with 4% paraformaldehyde then placed in a Shandon TBD-2 decalcifier (Thermo Fisher Scientific, USA) and frozen in isopentane chilled with dry ice before being embedded in Super Cryoembedding Medium (Section Lab, Japan). The samples were cut at −30 °C in an axial orientation into 7-μm slices with a CryoStar NX70 (Thermo Fisher Scientific, USA). Following standard protocols, Hematoxylin-Eosin-Safran and Alcian-Blue stainings were performed. Because IVD degeneration is well acknowledged to initiate mainly in the NP 30 , the IVDs were analyzed using modified Boos' scoring focusing only on the NP 31 . Cell density, mucous degeneration, tears and cleft formations, and granular changes were assessed to classify the IDD. The scores ranged from 0 (no degeneration) to 22 (complete degeneration). A blind evaluation of the histological samples was performed separately by two observers. If the scoring differed by less than two points, the median of the two values was used for the analysis. If the scoring differed by more than two points, the histological sample was examined once again by the two raters jointly to reach a consensus.
Statistical analyses. The statistical analyses were performed with R-software (4.0.3, The R Foundation, Vienna, Austria). Five measurements of the imaging data were collected for each MRI parameter with a 1-week interval between the measurements by two independent blinded observers.
• Intra-rater reliability for the Pfirrmann grading and each ROI on the T1, T2, and T2* mapping sequences: a linear mixed-effects model was used. The intra-rater reliability was considered to be good if the value was < 15% and moderate if the value was between 16 and 50%. • Inter-rater reliability for the Pfirrmann grading: the weighted kappa coefficient (κ) was used to determine the interobserver reliability for the Pfirrmann grading. The agreement was excellent for κ > 0.80, good for κ = 0.60-0.79, moderate for κ = 0.40-0.59, fair for κ = 0.20-0.39, and slight for κ < 0.2. The five readings were reviewed and the grade that was assigned the most to each IVD by each observer was retained for the determination of the inter-rater reliability. • Inter-rater reliability for each ROI on the T1, T2, and T2* mapping sequences: when the intra-rater reliability was < 15%, the mean of the five values was retained for the calculation of the inter-rater reliability. A coefficient of determination was used. Reliability was considered to be excellent if R 2 > 0.75, good if R 2 = 0.6-0.75, moderate if R 2 = 0.4-0.59, and poor if R 2 < 0.4. • Correlations between the mapping data and the Pfirrmann and the modified Boos' scorings were analyzed using Spearman's correlation. The correlations were considered significant if the p-value was < 0.05. For each MRI quantitative parameter, significant differences between the Pfirrmann grades were assessed using a linear mixed-effects model followed by comparisons between groups by a Wilcoxon-Mann-Whitney test. The differences were statistically significant when the p-value was < 0.05. For every linear mixed-effects model, the normality and the independence conditions of the residuals were validated according to the method described by Pinheiro and Bates 32 for mixed-effects models.

Results
Intra-observer agreement. The intra-observer analysis revealed strong agreement for all of the parameters (< 15%) except for T2*ROI2. The T2* relaxation time measurement for ROI 2 had moderate intra-observer agreement for observer 1 (20%). Detailed data are provided in Table S1 in the Appendix (Supplementary data).
As IVD regenerative therapeutic strategies are focused on the NP tissue and as the inter-observer agreement was variable for the AF ROIs, the decision was then made to focus the presentation of the results on the NP ROIs (ROIs 2, 3, 4, and B).

Imaging analysis.
No animal was excluded because of major vertebral disease. Five IVDs were counted in five sheep and six IVDs were counted in the five other sheep.
Regarding the Pfirrmann grading, 11 IVDs (22%) were classified as grade I, 17 (34%) as grade II, 16 (32%) as grade III, and 6 (12%) as grade IV. None were classified as grade V. The mean relaxation time values of the ten measurements for the different NP ROIs for each mapping method and for each Pfirrmann grade are summarized in Table 1   Mid-sagittal T2-weighted sequences and corresponding T1, T2, and T2* maps of lumbar ovine intervertebral discs with different Pfirrmann grades. Each pixel represents the quantitative relaxation time of T1, T2, or T2*. Pfirrmann grade I is identified by a homogeneous section that is isointense to the cerebrospinal fluid representing the Nucleus pulposus (NP), with a clear distinction between this structure and the Annulus fibrosus (AF). On the map images, the NP has a high relaxation time value, represented in red images on the color maps, while the AF, which is less hydrated, has low green relaxation time values. As the degenerative process involves a loss of the macromolecular structure of the disc and a decrease in its water content, the relaxation times in the NP and the AF decline. Pfirrmann grade IV is identified by an inhomogeneous central section, hypointense to the cerebrospinal fluid, with a loss of distinction between the NP and the AF and low relaxation time values in the AF and the degenerated NP, for which the pixels appear blue or green on the color maps. In terms of the T1 relaxation time measurements, statistically significant differences were found between Pfirrmann grades II and III (p < 0.001), I and III (p < 0.05), I and IV (p < 0.05), and between II and IV (p < 0.05) for all of the NP ROIs. By contrast, there was not a significant difference between grades I and II and between grades III and IV for any of the four NP T1 ROIs (Table 2a).
Analysis of the T2 relaxation time measurements revealed statistically significant differences between all of the Pfirrmann grades except between grades I and III for T2 ROIs 2, 3, and B. Only T2 ROI 4 had significant differences between all of the Pfirrmann grades (Table 2b).
Similarly, for the T2* relaxation time measurements, statistically significant differences were found between all of the Pfirrmann grades except between grades I and III for T2* ROIs 2, 3, and B. For T2 ROI 4, there were statistically significant differences between all of the Pfirrmann grades except between grades I and II (Table 2c).
Correlation between the MRI relaxation time measurements and the modified Boos' scoring. The histological scores ranged from three for the healthiest IVD to nineteen for the most degenerated one. The mean value was 9.4 and the median value was 9. Graphical distribution of the histological scores of the IVDs is provided in Figure S2 in the Appendix (Supplementary data).
The Spearman correlations showed that the modified Boos' scoring was negatively correlated with the NP values measured for ROIs 2, 3, 4, and B on the T1 mapping images (Fig. 3), the T2 mapping images (Fig. 4), and the T2* mapping images (

Comparison between the two methods of drawing ROIs for clinical application.
Considering the ROI B from the 3-ROIs method and the three NP ROIs from the 5-ROIs method together, we found that the two methods were quite similar for evaluation of the NP relaxation time values on T1, T2, and T2* mapping images.
Considering the time required for each method, the mean time for the 3-ROIs method was much faster than the 5-ROIs method (50 s vs. 120 s). These times do not take into account the selection of the two sequences and their correct display on the screen, which do not vary from one method to the other. The number of computer manipulations was much higher for the 5-ROIs method compared to the 3-ROIs approach (22 vs. 11).

Pfirrmann score probabilities.
With an ordinal regression, the probability of the degenerative degree according to the relaxation time value was obtained for T1 ROI B, T2 ROI B, and T2* ROI B and are illustrated in Fig. 6.
Considering ROI B, the probability of a non-degenerated NP (Pfirrmann grade I) or an NP with few degenerative changes (Pfirrmann grade II) is higher than 80% if the T1, T2, and T2* relaxation times are greater than 1,320 ms, 164 ms, and 59 ms, respectively.

Discussion
This study investigated the effectiveness and the feasibility of MR T1, T2 and T2* mapping MRI using two drawing ROI methods on the ovine lumbar IVD. These sequences were correlated to qualitative MR Pfirrmann grade and histological scoring in order to find the most suitable approach that can readily be used in practice in veterinary and human medicine. In accordance with previous studies, most of the AF ROIs offer poor-to-moderate inter-agreement, contrary to the NP ROIs, irrespective of the MRI quantitative sequences 27,33 . The manually  A strong negative correlation has been found between T2 and T2* times and MR qualitative Pfirrmann grading of IDD, which is consistent with our results 10,12,20,[33][34][35][36][37] . In contrast, T1 mapping has rarely been used to assess IDD, although it has been studied extensively for other tissues' perfusion such as brain or myocardium 38 . Very few studies have applied T1 mapping sequences to the IVD, all of which used the multiple-inversion-recovery method 28,36,39,40 . Our study, to the best of our knowledge, is the first to utilize the variable-flip-angle method and to demonstrate the strong correlation between qualitative MR and histology. The main difference between the two acquisition methods is the time of acquisition of the sequences. The number of inversion-recuperation sequences varies between 5 and 16. Only one of those studies provides the actual time of acquisition of each www.nature.com/scientificreports/ sequence 28 , in which the total time of acquisition of the T1 mapping was 784 s, while our method only took 88 s, thus allowing for time savings in daily practice. Relaxation time measurement is a real challenge in terms of the risk of bias. In this study, great care was taken to reproduce the MR acquisitions under similar conditions. Diurnal significant variation has been highlighted previously in the NP using T2 41,42 and T1 mapping 28 in healthy human lumbar IVD due to the decrease in water content that occurs during a diurnal cycle. One study has assessed this variation in the human IVD using T2* mapping but no significant differences have been identified (p-value = 0.748) 43 . In sheep, a potential diurnal variation in mapping values has not been investigated yet but, for ethical considerations, we decided not to anesthetize our cohort twice in one day to explore this feature and realized all the MR exams in the morning. In in vivo studies, quantitative MRI is also sensible to motion artifacts induced by respiration and involuntary displacements of the specimen during the acquisition of the sequences 39 but the fact that our study was conducted on anesthetized sheep put under mechanical ventilation limited the presence of these artifacts. Moreover, mapping sequences may be prone to artifacts. When selecting the FOV of each mapping sequence, particular care was taken to select not only the lumbar IVDs but to extend to the last thoracic IVDs and the lumbosacral junction. This allowed spatial warping at the border of the images to be avoided and thus limit such artifacts.
As there is no standardized method for ROI acquisition, comparison of mapping data is not straightforward. Numerous approaches are used to assess IVD degeneration that focus on various parts of the IVD. In accordance with the study that compared three ROI drawing methods 27 , our data confirm the moderate-to-poor interobserver agreement for AF ROIs, unlike NP ROIs. No pertinent difference was found between the two approaches used in our study to define the NP and its junction with the AF. Although ROI 4 appears to be more precise at distinguishing the degenerative grade, especially on T2 images, the drawing method to obtain this ROI implies drawing the 5 ROIs together before isolating the mean value obtained with ROI 4. On the other hand, in the 3-ROIs method, the ROIs are independent and can be drawn individually. In clinical practice, the mean time required to obtain ROI 4 is then greater than for ROI B (150 vs. < 50 s). Moreover, the 5-ROIs method appears to be more complicated than the other method in terms of software manipulations and thus involves a steeper www.nature.com/scientificreports/ learning curve, especially for an untrained observer. Therefore, of the three mapping methods, ROI B appears to be the most suitable tool for evaluation of the NP, and it warrants being assessed for routine clinical practice in human medicine. ROI B has many benefits: ease of drawing, timeliness, and sensitivity. Even if a perfect match with human IVD degeneration can't be found, animal models are crucial in studies related to degeneration or regeneration therapeutic strategies of this tissue, and, notably, spontaneous degenerative models 23,24,26 . In particular, ovine lumbar spine and discs present a lot of similarities with human tissues regarding the anatomy, biology, and biomechanics [44][45][46] . Preclinical investigations with intradiscal injections are essential to test the new therapeutics developed in vitro before clinical transposition to human patients. Surprisingly, if sheep have been well described in IVDD preclinical studies, very few investigations using quantitative MRI have been conducted on this species. T2* mapping and T2 mapping have been negatively correlated with disc degeneration using modified Boos 19 and non-enzymatic glycation 47 , respectively. T2 reference values according to the sheep age have been established in a study that used MRI and CT to assess the spontaneous disc degeneration 25 . Our data indicated that T1, T2, and T2* mapping are applicable in sheep to estimate the changes occurring in the NP in preclinical regenerative studies focusing on this tissue.
Finally, few studies have assessed the correlation between in vivo quantitative MR measurements and histology of the NP. Investigations using animal models facilitate the comparison of imaging modalities and histology. Boos' scoring has been used in a T2* mapping study of the lumbar ovine NP of healthy IVDs, but not for correlation 19 . Our study is the first, to our knowledge, to demonstrate the correlation between T1, T2, and T2* mapping and histology in an ovine model of lumbar IVD degeneration. Imaging may be a valuable substitute for histological analysis of the ovine IVD, by limiting the number of animals sacrificed in accordance with the The mean T2* relaxation time of the NP was measured by regions-of-interest (ROIs) that included the NP or a part of the NP of each lumbar ovine disc on the sagittal T2-weighted image and by copying this ROI onto the corresponding T2* mapping image. Equal-oblong ROIs delimitated (a) the junction between the NP and the ventral Annulus fibrosus (AF) (T2* ROI 2), (b) the NP (T2* ROI 3), and (c) the junction between the NP and the dorsal AF (T2* ROI 4). (d) A manually drawn oval delineated the NP (T2* ROI B). The modified Boos' scoring was negatively correlated with the NP values measured for ROIs 2, 3, 4, and B on the T2* mapping images (p < 0.001).

Scientific Reports
| (2022) 12:5398 | https://doi.org/10.1038/s41598-022-09348-w www.nature.com/scientificreports/ 3R-ethics-rules. Moreover, longitudinal assessment of the IVD is likewise useful to study the changes that occur in early IDD. However, while T2 and T2* mapping have been correlated to the water content and molecular interactions 11,14 and to the spatial architecture of the macromolecules, respectively, further studies are needed to establish which molecular changes the T1 value reflects in the IDD process.
In conclusion, quantitative T2, T2*, and T1 mapping using a single manual oval ROI focused on the NP are easy, reliable, time-efficient, and reproducible tools to evaluate lumbar IVD degenerative disease in sheep and they could be useful tools for routine clinical practice. Early detection and characterization of IVD degeneration will need such tools to assist with the diagnosis and to help with the therapeutic decision. In particular, variableflip-angle T1 mapping appears to be a relevant modality for assessment of the pathological changes in the lumbar IVDs and it warrants particular consideration in further preclinical and clinical studies.