Three-dimensional meniscus allograft sizing—a study of 280 healthy menisci

Background Inaccurate meniscus allograft size is still an important problem of the currently used sizing methods. The purpose of this study was to evaluate a new three-dimensional (3D) meniscus-sizing method to increase the accuracy of the selected allografts. Methods 3D triangular surface models were generated from 280 menisci based on 50 bilateral and 40 unilateral knee joint magnetic resonance imaging (MRI) scans. These models served as an imaginary meniscus allograft tissue bank. Meniscus sizing and allograft selection was simulated for all 50 bilateral knee joints by (1) the closest mean surface distance (MeSD) (3D-MRI sizing with contralateral meniscus), (2) the smallest meniscal width/length difference in MRI (2D-MRI sizing with contralateral meniscus), and (3) conventional radiography as proposed by Pollard (2D-radiograph (RX) sizing with ipsilateral tibia plateau). 3D shape and meniscal width, length, and height were compared between the original meniscus and the selected meniscus using the three sizing methods. Results Allograft selection by MeSD (3D MRI) was superior for all measurement parameters. In particular, the 3D shape was significantly improved (p < 0.001), while the mean differences in meniscal width, length, and height were only slightly better than the allograft selected by the other methods. Outliers were reduced by up to 55% (vs. 2D MRI) and 83% (vs. 2D RX) for the medial meniscus and 39% (vs. 2D MRI) and 56% (vs. 2D RX) for the lateral meniscus. Conclusion 3D-MRI sizing by MeSD using the contralateral meniscus as a reconstruction template can significantly improve meniscus allograft selection. Sizing using conventional radiography should probably not be recommended. Trial registration Kantonale Ethikkommission Zürich had given the approval for the study (BASEC-No. 2018-00856).


Background
The meniscus plays an important role in the kinematics of the knee joint, reduces contact pressure [1], and improves joint stability [2,3]. These functions disappear after subtotal or total meniscectomy with resultant early osteoarthritis [4].
3D meniscus sizing was recently proposed as a solution to increase the accuracy and precision of meniscus allograft selection [40]. The contralateral meniscus can be used as a very precise reconstruction template. However, advantages of 3D-MRI sizing with the contralateral meniscus as a reconstruction template compared with existing sizing methods have not been shown to date. We hypothesized that 3D-MRI sizing with the contralateral side is clearly superior compared with the currently used sizing methods.

Methods
The following description of the approaches used for the validation of the different sizing methods is divided into three parts. Part 1 focuses on describing the basic material, imaging, and measurement methods. Part 2 is dedicated to explaining the three different sizing methods in detail. Finally, part 3 contains the sizing validation methods.

Part 1: Material/imaging/measurements Material
Fifty patients with bilateral and 40 patients with unilateral complete imaging (34 men, 56 women; mean age 26.7 years (range 15-50)) were retrospectively included in this study. All patients had a patellofemoral disorder. The inclusion criteria were patients with available MRI scan and conventional radiography, mature skeletal age, healthy contralateral meniscus, and no tibio-femoral osteoarthritis (Kellgren and Lawrence grade 0) [41]. The presence of a completely closed growth plate at the distal femur and the proximal tibia in the MRI images was used to determine mature skeletal age. Meniscus integrity was assessed by a fellowship-trained musculoskeletal radiologist, and the presence of meniscus tears, degeneration, or extrusion led to exclusion.

Imaging
All radiographs (RX) were performed in a standard fashion, with a plain anteroposterior (AP) and lateral (LAT) view with the ampoule placed 1 m distant from the knee using calibrators for the correction of magnification (Optimus 50 X-ray Generator; Philips, USA).
All MRI scans were performed in our institution on a 3.0-T scanner (Skyra-fit, Siemens Healthineers, Erlangen, Germany) with a send/receive knee coil, and the patient was examined in the supine position. All MRI examinations consisted of sagittal, coronal, and axial sequences as part of the standard MRI protocol [40]: (1)

Measurement methods
The complete dataset was further analyzed based on the following models and measurement methods: 1) 3D surface model of medial and lateral meniscus; 2) 3D-calculated meniscal width, length, and height; 3) 2D-calculated meniscal width and length in MRI; and 4) 2D-derived meniscal width and length in RX. 1) 3D surface model of medial and lateral meniscus 3D surface models were created, as described in a previous paper [40], using the Materialise Interactive Medical Control System (MIMICS) 18.0 3D reconstruction software program (Materialise, Leuven, Belgium). The segmentation of the medial and lateral meniscus was manually performed by two trained orthopedic surgeons in sagittal and coronal slides (biplanar). Finally, all 3D surface models were smoothed (gap closing distance 0.0 mm, smallest detail 1.0 mm), and all left menisci were mirrored to the right surface models.
2) 3D-calculated meniscal width, length, and height ( Fig. 1a) The 3D measurement of meniscal width, length, and height was performed in the same way as described in a previous paper [40] with an oriented bounding box (= minimal-volume rectangular box that fully encloses the meniscal model) using the in-house developed planning software CASPA (Computed Assisted Surgery Planning Application, Balgrist CARD AG). The box was initially automatically generated around the 3D models. Then, the box was aligned parallel to a line from the anterior to the posterior meniscus root by rotation around the zaxis. Finally, the box dimensions were adjusted as long as the entire meniscus body was enclosed. Meniscus width (y-axis), length (x-axis), and height (z-axis) were now represented by the dimensions of the bounding box.
3) 2D-calculated width and length in MRI (Fig. 1b) For the measurement of the 2D meniscal width and length in MRI, we used the following method. The measurements were performed on axial slides with reference to sagittal slides. Anterior and posterior meniscus roots were identified and connected by a first line. Parallel to this, a second line was placed adjacent to the outer contour of the meniscal base. Two additional lines, perpendicular to the first two lines, were set on the outer contours of the meniscus. Thereby, a 2D bounding box was created. The width and length of the meniscus corresponded to the dimensions of the box and could be measured. (Fig. 1c) The tibial plateau width was measured in the AP view perpendicular to the joint line as the distance between the margin of the tibial metaphysis to the medial lateral tibial eminence. According to Pollard, meniscal width is supposed to be equal to the tibia plateau width for medial and lateral menisci [31]. Tibia plateau length was measured in the LAT view perpendicular to the joint line as the distance between the anterior surface of the tibia above the tuberosity to the posterior margin of the tibia plateau. According to Pollard, the medial and lateral meniscal lengths can be calculated as 80% and 70%, respectively, of the tibia plateau length (medial meniscal length = 0.8 × tibia plateau length; lateral meniscal length = 0.7 × tibia plateau length).

Part 2: Sizing methods
For the 3D-MRI meniscus-sizing method and for the 2D-MRI meniscus-sizing method, we used the contralateral meniscus as a template [40]. Therefore, both sizing methods are direct sizing methods, i.e., they were compared with the size of the meniscus on the contralateral side. For the 2D-RX sizing method, sizing was performed using an indirect calculation based on the bony anatomy of the same side [31].

1) 3D-MRI meniscus sizing (Fig. 2)
Principle: Direct meniscus sizing, based on the mean surface distance (MeSD) of the contralateral meniscus. Requirements: MRI scan of the healthy contralateral knee and 3D surface models of the allografts. Surface distance: The similarity of two objects can be represented by the closest surface point distance as previously described [40]. To this end, the objects were automatically superimposed by using the iterative closest point (ICP) algorithm so that the distances of all surface points from one object to the other (and vice versa) were as small as possible. The maximum surface distance (MaSD) is the widest measured distance, and the MeSD is the mean of all surface distances. For this analysis, the Hausdorff distance, the highest distance value between the two objects, was used (Fig. 3). Allograft selection: The best fitting allografts were selected based on the closest MeSD values between the 3D template and all available 3D allograft surface models.

2) 2D-MRI meniscus sizing
Principle: Direct meniscus sizing, based on calculated meniscal width and length of the contralateral meniscus. Requirements: MRI scan of the healthy contralateral knee, and the measured width and length of the allografts. Allograft selection: The best fitting allografts were selected by the smallest differences in width and length between the calculated sizes and the available allografts.

3) 2D-RX meniscus sizing
Principle: Indirect meniscus sizing, based on derived meniscal width and length calculated from the measurement of the ipsilateral tibia plateau according to Pollard [31]. Requirements: Conventional radiography of AP and mediolateral view of the ipsilateral side with calibrators. Allograft selection: The best fitting allografts were selected by the smallest differences in width and length between the calculated sizes and the available allografts.

Part 3: Sizing validation
First, different terms must be clearly defined for better understanding. The missing meniscus-i.e., after meniscectomy-is hereby called the "original meniscus." The 3D meniscus surface model for meniscus sizing is hereby called the "template meniscus." The best fitting meniscus that was selected by a particular sizing method is hereby called the "selected meniscus." In other words, we used different sizing methods of the template meniscus to choose the selected meniscus and then compared the selected meniscus with the original meniscus.
Because sizing with surface distances does not allow a simple comparison of two values (i.e., width and length) of the original and selected meniscus, as was done in previous studies, sizing validation had to be performed by a more complex simulation (see Fig. 4). Therefore, the 50 patients with bilateral imaging served as the "validation group," resulting in 100 (= 50 right and 50 left) different medial and lateral menisci. An imaginary tissue bank ("allograft pool") was composed of all 100 menisci, together with the menisci from the unilateral imaging of another 40 patients, resulting in 140 different menisci. The menisci of each tested patient in the validation group were excluded from the potential menisci that could be selected for them, which finally resulted in an  Surface distance (SD). The SD between two 3D models can be calculated after automatic superimposition by using the iterative closest point algorithm so that the distances of all surface points from one model to the other (and vice versa) are as small as possible. Maximum SD (MaSD) is calculated as the widest distance between "meniscus A" and "meniscus B." The mean SD (MeSD) is the mean value of all surface points between "meniscus A" and "meniscus B." For this analysis, the Hausdorff distance, i.e., the highest values, is used allograft pool of 138 different medial and lateral menisci. Therefore, for the validation group, we needed bilateral MRI scans to compare the previously described sizing methods. For the additional patients, this was not necessary, and we aimed to have as many allografts as possible in our tissue bank.
The simulation was then performed for each meniscus (meniscus No. 1 to meniscus No. 100) of the validation group. An in-house developed computer program automatically calculated the closest MeSD and MaSD between all 100 original (validation group) and 138 different allografts (allograft pool). Overall, there were 13,800 possibilities for allograft selection by MeSD (= 100 originals × 138 allografts), and the 100 closest menisci served as the best possible allografts, which served as the gold standard. In a second step, the same validation was now performed by the contralateral meniscus surface model as the template, which corresponds to the 3D-MRI sizing. Of the 13,800 possibilities, only the best fitting allografts by this sizing method were selected. In a third step, validation was performed using the closest difference in measured width and length of the contralateral side in MRI (2D-MRI sizing) and 3D measured width and length. Thereby, width and length were equally weighted and selected by the lowest error sum of squares: (width sized −width allograft ) 2 + (length sized −length allograft ) 2 . In a fourth step, the same procedure was repeated based on the closest difference in derived width and length of the ipsilateral side (2D-RX sizing).
Finally, the original meniscus was compared with the selected menisci resulting from the different sizing methods using the surface distance as well as the meniscal width, length, and height as similarity measurements.
Meniscus outliers were defined as those with a difference between the original and selected meniscus over 5 mm in width, over 5 mm in length, or over 4 mm in height.

Inter-rater/intra-rater reliability
The first 30 knee joints-in alphabetic order based on patient names-were used to calculate interclass correlation. 3D surface models (3D-MRI sizing), as well as width/length in MRI (2D-MRI sizing) and X-ray (2D-RX sizing), were calculated by two trained orthopedic surgeons, as described above. Intra-rater reliability was calculated by the repetition of these measurements by one of the surgeons.

Statistics
The statistical analysis was performed with SPSS (IBM Corp. IBM SPSS Statistics for Windows, version 24.0. Armonk, NY: IBM Corp.) Descriptive analyses and independent-sample t tests were performed to investigate the meniscal diversity. P values below 0.05 were considered statistically significant. The correlations between width/length and MeSD/MaSD were analyzed with a linear regression analysis (Pearson correlation coefficient). Inter-rater and intra-rater reliability was calculated using interclass correlation coefficients (ICCs) (two-way mixed, absolute agreement). We used scatterplots and boxplots for visual presentation. . The 3D surface model of the contralateral side corresponded to the 3D-MRI sizing. The meniscal width/length of the contralateral side in MRI corresponded to the 2D-MRI sizing. The derived meniscal width/length by the ipsilateral tibia plateau in RX corresponded to the 2D-RX sizing method. The best match ("selected meniscus") out of the "allograft pool" was selected based on the different "template menisci." Finally, the "selected meniscus" was compared with the "original meniscus." This procedure was repeated for all 100 menisci of the "validation group"

Sizing validation
All results are summarized in Tables 1 and 2  None of the presented sizing methods always selected the best available fitting allograft (which corresponded to the original meniscus) out of the imaginary tissue bank. In particular, the 3D shape (MeSD and MaSD) in these cases was different between the selected and the original meniscus, while the width, length, and height were not distinct.
Furthermore, 3D-MRI sizing reduced the number of outliers. For the medial meniscus, there were 12% fewer outliers selected compared with 2D-MRI sizing and 48% fewer outliers selected compared with 2D-RX sizing. For the lateral meniscus, there were 7% fewer outliers selected compared with 2D-MRI sizing and 14% fewer outliers selected compared with 2D-RX sizing.

Discussion
The most appropriate methods for meniscus sizing are still debated. Even under laboratory conditions, the most familiar radiographic method of Pollard had a standard error of prediction of ± 2.9 mm for meniscal width and ± 3.8 mm for meniscal length [31], which corresponds to a standard error of approximately 8%. Considering that the proposed 10% graft mismatch could cause relevant articular problems (i.e., 3-or 4-mm size difference based on meniscal sizes reported in the literature), the Pollard method seems to not be completely safe [17,37]. Although MRI sizing seems to be superior to the radiological methods, 17% of the measured menisci in a cadaver study by Shaffer had differences of over 5 mm compared with the anatomical measurements [36].
Other authors have shown better results with an average error rate of 4.0-5.3% [35]. Because sizing is performed based on two variables (i.e., width and length), inaccuracy is probably magnified by the combination of both. In a previous study, a new 3D meniscus sizing method was proposed, which was based on a healthy contralateral meniscus template to reduce inaccuracy [40]. For this sizing method, there are some basic prerequisites, which must be fulfilled and will be discussed in the following section. First, an easily determined, poseinvariant measurement parameter is required to accurately compare freely moveable 3D bodies. Second, the segmented 3D meniscus has to correspond with the actual 3D anatomical shape. Third, the contralateral meniscus has to be similar. Fourth, the benefits of a 3D sizing method must be compared with those of the currently used sizing methods. Fifth, the additional benefit has to be validated with improved clinical results.
Basic prerequisites for 3D sizing 1) The similarity of 3D models can be evaluated by the surface distance, as already described in a previous study [40]. However, it is not yet known whether 3D sizing should be performed by the closest MeSD or by the closest MaSD, or a combination of these measures. Depending on the chosen criterion, the best allograft would be the selected according to the MeSD or to MaSD. Because there were poor correlations between these two values (R 2 = 0.430-0.495), similar to the poor correlations between meniscal width and length (R 2 = 304-310), these values could provide different information. The MeSD was generally less variable as the MaSD could be affected by one single surface point with a wide distance between the two models-for example, due to a partly segmented transverse or meniscotibial ligament. Therefore, we assessed sizing primarily based on the MeSD. segmented shapes remains unavailable. However, to date, MRI measurements are often used as the gold standard and seem very good at identifying the meniscus tissue [32,33,35,37,39]. 3) In a previous study, we showed that the contralateral side can be used as a very precise and reliable meniscus template for 3D sizing. 4) Our results are not directly comparable to previous studies. 3D sizing is based on MeSD and not on meniscal width and length. Therefore, a comparison to previous studies was not possible. We solved this problem by a simulated allograft selection based on the three sizing methods from an imaginary tissue bank of 138 different menisci. The selected allografts could be easily compared. An important advantage of this evaluation was the fact that the multiplicative inaccuracy due to a combination of two imprecise parameters (i.e., width and length) was taken into account. Herein, we showed a significant improvement of the 3D shape of the selected meniscus allograft with sizing by the MeSD compared with sizing by the combined meniscal width and length. There were also improvements based on meniscal width, length, and height measures, but these improvements were not statistically significant for all values. Similar to many other studies, the mean width and length values of the selected menisci were good and mostly acceptable across the three sizing methods. However, the main problem of meniscus allograft sizing is related to the number of selected outliers. In our opinion, this is exactly the strength of the 3D sizing method. The number of outliers selected could be reduced by 12-29% compared with 2D-MRI sizing and by 41-53% compared with 2D-RX sizing. We defined outliers as a size difference between the original and selected meniscus of > 5 mm in width or length and > 4 mm in height. A cutoff value of 5 mm for width and length has previously been used by other authors [25,36]. A cutoff value based on meniscal height had not been previously described. In our collection, the meniscal height had a mean value of 9.3-9.9 mm with a standard deviation of only ± 1.4 mm. Therefore, a cutoff value for meniscal height of > 5 mm would probably never become applicable. Because the clinical evidence regarding meniscal height is still unclear, we did not choose our cutoff value based on the 10% meniscal mismatch rule [17], but on > 4 mm, which was slightly higher than the slide thickness and detected only 5 mismatched menisci overall.
There are few limitations of this study. As already mentioned under the second point above, a direct comparison of the anatomical and segmented shapes is missing. However, the meniscus can be accurately identified in MRI, and MRI is therefore often used as the gold standard [32,33,35,37,39]. Because a mismatch of 3-4 mm (10%) appears to be clinically relevant, a slide thickness of up to 3 mm can significantly influence the results. Therefore, a biplanar segmentation of the meniscus with excellent inter-rater and intrarater reliability for width and length (ICC inter 0.913-0.973; ICC intra 0.955-0.987) was used [40]. Furthermore, a larger tissue bank could probably have improved our results. However, the number of allografts used is probably sufficient, as there are many tissue banks with (much) less than 138 different available allografts (on request).
3D meniscus sizing has some great advantages, as already mentioned in a previous study [40]. In the present study, we demonstrated-even with a limited number of 138 allografts-that 3D sizing can significantly improve the accuracy and substantially reduce the number of relevant outliers. In conclusion, the limited number of meniscus allografts due to increased demand in the world could be much better matched to the patients. Moreover, this method can be used to do more than find the best fitting 3D allograft. If necessary, a slightly smaller size due to presumed extrusion or slightly greater size due to presumed shrinkage can be easily found by down-or upscaling the 3D template [14]. Furthermore, the feasibility of "bone plugs" or "bone bridge" fixation methods, which are supposed to be superior to "soft-tissue" and "suture bone tunnel" fixation techniques, could also be improved due to the 3D method identifying the shape of the meniscus [14,42]. This 3D template could also be used in the future for biomimetic 3D-printed scaffolds [43,44].
However, 3D sizing involves additional costs. An MRI scan of the contralateral side is needed, and the tissue banks must be willing to offer this option. 3D surface models must be generated for all allografts. In the future, meniscus segmentation could probably be performed in a semiautomated or fully automated manner [45,46], which could reduce additional costs. Automatized comparison of the meniscus template of the healthy contralateral meniscus and all available allografts is already possible.

Conclusion
3D-MRI sizing by MeSD using the contralateral meniscus as a reconstruction template can significantly improve the accuracy of meniscus allograft selection and reduce outliers compared with sizing that relies only on width and length in radiography and MRI. Using these methods, the limited number of available allografts could probably be more effectively distributed based on the rising demand in the world.