MRI-Based Modeling for Radiocarpal Joint Mechanics: Validation Criteria and Results for Four Specimen-Specific Models

Short abstract

The objective of this study was to validate the MRI-based joint contact modeling methodology in the radiocarpal joints by comparison of model results with invasive specimen-specific radiocarpal contact measurements from four cadaver experiments. We used a single validation criterion for multiple outcome measures to characterize the utility and overall validity of the modeling approach. For each experiment, a Pressurex film and a Tekscan sensor were sequentially placed into the radiocarpal joints during simulated grasp. Computer models were constructed based on MRI visualization of the cadaver specimens without load. Images were also acquired during the loaded configuration used with the direct experimental measurements. Geometric surface models of the radius, scaphoid and lunate (including cartilage) were constructed from the images acquired without the load. The carpal bone motions from the unloaded state to the loaded state were determined using a series of 3D image registrations. Cartilage thickness was assumed uniform at 1.0 mm with an effective compressive modulus of 4 MPa. Validation was based on experimental versus model contact area, contact force, average contact pressure and peak contact pressure for the radioscaphoid and radiolunate articulations. Contact area was also measured directly from images acquired under load and compared to the experimental and model data. Qualitatively, there was good correspondence between the MRI-based model data and experimental data, with consistent relative size, shape and location of radioscaphoid and radiolunate contact regions. Quantitative data from the model generally compared well with the experimental data for all specimens. Contact area from the MRI-based model was very similar to the contact area measured directly from the images. For all outcome measures except average and peak pressures, at least two specimen models met the validation criteria with respect to experimental measurements for both articulations. Only the model for one specimen met the validation criteria for average and peak pressure of both articulations; however the experimental measures for peak pressure also exhibited high variability. MRI-based modeling can reliably be used for evaluating the contact area and contact force with similar confidence as in currently available experimental techniques. Average contact pressure, and peak contact pressure were more variable from all measurement techniques, and these measures from MRI-based modeling should be used with some caution.

Introduction

With an aging U.S. population, the need for cost-effective tools for the detection and prevention of osteoarthritis (OA) is increasing. Joint mechanics are very important for understanding OA, as OA has been associated with abnormal joint kinematics and/or loading [1–8]. However, the role of mechanical factors in the development of OA, specifically the levels of abnormal joint loading that will trigger OA, are not known. The development of diagnostic tools has been an important area of investigation for both clinicians and biomechanical engineers.

Magnetic resonance imaging (MRI) has the demonstrated capability for the early detection of OA, even prior to clinical symptoms. Previous studies of detection have focused on MRI protocols that produce measures of proteoglycan (PG) content. These protocols include T2 relaxation time mapping, T1ρ mapping, dGEMRIC (delayed gadolinium enhanced MRI of cartilage) T1 mapping, and sodium mapping, and some studies have additionally correlated these measures to mechanical properties of the cartilage [9–17]. While such techniques may soon be clinically applicable for identifying early OA, they cannot detect conditions that will lead to OA before any degeneration occurs. Thus, it would be further beneficial to identify joint mechanics that will lead to OA, so that a correction can be made to prevent OA. While many investigations have used computational (primarily finite element) models to assess joint mechanics, relatively few have done so using in vivo subject-specific geometry and kinematics [18–24]. Thus, we have sought to develop models based on MRI of human subjects during functional loading, an approach that would minimize assumptions considered in the modeling process.

Pillai et al. proposed and illustrated a method for modeling radiocarpal joint contact mechanics based on MRI data collected during functional loading [25]. MR images of the wrist of four human subjects were acquired in both grasp and relaxed states, using a 1.5 T whole body MRI scanner with a flexible wrist coil. Image data was used to generate surface models and to determine bone positions and orientations for the loaded state. The geometric models and kinematic data were then implemented in the Joint_Model program developed at Columbia University to investigate joint contact mechanics [26]. The contact pressure distribution, contact forces, and contact areas were analyzed for each of the articulations of the radiocarpal joint (radioscaphoid, radiolunate, and scapholunate). The results of that study were promising, although the image resolution was low and the approach was not validated. We began to address the shortcomings in that feasibility study by improving our images and refining the joint contact modeling in a preliminary validation of the method’s accuracy [27]. Specifically, we increased the image resolution and signal-to-noise ratio with the use of a gradient recall echo sequence (in-plane resolution to 0.12 mm and a slice thickness of 1 mm) in a 9.4 T high field MR scanner. Three cadaveric forearm specimens were tested for radiocarpal contact using both a Pressurex pressure sensitive film and a Tekscan piezoelectric sensor in the joint. Light grasp was simulated by applying static forces to the flexor digitorum profundus (FDP), flexor digitorum superificialis (FDS), and flexor pollicis longus (FPS) tendons. The results were encouraging—peak pressures were within acceptable values, and the interpenetration depth of the models did not exceed realistic values. The values from direct calculation of contact areas (from the images) closely matched the contact areas estimated by the refined model.

The current study was undertaken to extend the validation of our MR imaging and modeling approaches with tests performed on an additional four specimens. We also implemented improved experimental and modeling techniques, and incorporated multiple validation criteria in our analysis. Our measures were considered to have met validation criterion when the model data fell within two times the accuracy of the experimental method [28]. In the following sections, we describe the details of our approaches and discuss the findings that resulted from our investigations.

Methods

Specimen Preparation and Experimental Setup.

Four unembalmed human cadaveric forearms were collected and kept frozen until use. The dissection process consisted of two primary steps. First, the required tendons and bones were exposed and isolated. This was followed by volar dissection of the radiocarpal joint capsule.

During dissection, three flexor tendons were isolated for simulated grasp loading: the flexor digitorum superficialis (FDS), the flexor digitorum profundus (FDP), and the flexor pollicis longus (FPL). Two extensor tendons, the extensor carpi ulnaris (ECU) and extensor carpi radialis (ECR, the combined ECR brevis and longus tendons) were also isolated (Fig. 1) for tendon loading to stabilize the wrist and prevent wrist flexion. Suture loops were added to all tendons to facilitate loading. The suture loops were secured to nylon monofilament, which was connected to weights hanging over a pulley system to simulate muscle contraction. By loading the extensors along with the flexor tendons, a more physiological simulation of light grasp was performed. Other than the 5 tendons and the interosseous membrane of the forearm, all remaining tissue along a 10 cm section of the radius and ulna (from about 4–14 cm proximal to the wrist) was removed. Two 6.35 mm holes were drilled through the radius, as well as one through the ulna. Plastic bolts were used to attach the specimen to a plastic mounting plate (Fig. 1). Care was taken to secure the bones such that the forearm was in neutral rotation.

Fig. 1.

Isolation and suturing of the tendons and fixation of radius and ulna to PVC plate. The radius has two plastic bolts and ulna has one. The isolated and sutured extensor carpi radialis tendons are clearly seen at the bottom of the image.

All materials used in the experimental setup were made of plastic so as to be compatible with the high-field MRI scanner. Calibrated plastic water jugs constituted the weights used to apply a total of 110 N across the wrist. The flexor loads were based grossly on the cross-sectional area of each muscle, and the wrist extensor loads were determined by trial and error to assure the wrist did not flex. Loads of 30 N each were first applied to the ECU and ECR to stabilize the wrist against wrist flexion. Loads of 20 N each were applied to the FDS and FDP, and 10 N was applied to the FPL to simulate grasp. The same loads were applied during physical measurements and during MRI scanning. During load application, a short section of 1.5 in. (37 mm) outside diameter PVC pipe was placed in the fingers of the specimen to simulate grasp. The digits were taped over the pipe to help maintain a consistent position between loading trials.

The specimen was also dissected on the volar aspect of the wrist [27]. A skin and subcutaneous tissue incision was made along the radiocarpal border, as determined by palpation. Upon further blunt dissection, the volar joint capsule (and associated ligaments) was released from the radial styloid to the ulnar edge of the radius. This exposure allowed the insertion of pressure sensors during the experiment. Because wrist ligaments had to be cut to insert the film into the joint, the carpal kinematics may have deviated from normal physiological motions, and the overall radiocarpal mechanics could no longer be considered completely physiological. The main concern; however, was that the radiocarpal mechanics are repeatable between trials and across measurement systems.

Experimental Pressure Measurements.

The two primary methods of measuring joint contact pressure distributions are Fuji-scale pressure sensitive (dye release) film and thin film electronic pressure sensors. Because of the relative advantages and limitations of each experimental technology, this study used both methods and compared the results of each to data from MRI-based joint contact models.

The Fuji-scale film used was Super Low Pressurex^® film (Sensor Products Inc., East Hanover, NJ), with a standard measurement range of about 0.5–2.5 MPa (70–350 psi). Super Low was a two part film (sensor/developer) with a total thickness of 3 mil (0.08 mm), and the color intensity was directly related to the local applied pressure. Though the manufacturer provided a color chart for reading the pressure, we performed same day, same sheet 10-level calibration tests using a materials testing machine (Instron, Model # 5811, Instron Corp., Canton, MA) and analyzed the normalized red minus normalized blue color content for the most sensitive measure of pressure [29]. The limitations of this film were the threshold for recording data, the possibility of saturating the film with pressures outside the design range, and (perhaps most notably) the possibility of artifacts during insertion and removal of the film.

On the basis of the shape of the radial fossae from prior specimens, a template was created with the shape of the two articular surfaces on the radius. The Pressurex film was cut to match the shape of the template to facilitate insertion of the film and reduce buckling artifacts. In addition, a small volar tab was added to the film, to provide a place to grip it during insertion and removal. The tab also served as a permanent reference to the orientation of the film in the joint. The film and developer sheets were wrapped in a thin layer of plastic to prevent saline solution and/or synovial fluid from inhibiting development of the color on the film. The film was then inserted into the joint and loads were applied. Specifically, the ECU and ECR loads were applied, then FDS and FDP, and finally the FPL. This order was repeated for all experiments and specimens. The load was held for approximately 30 s then released, and the film was carefully removed. This procedure was repeated five times per specimen, to assure representative measurements with minimal artifact. Because of the tendency for artifacts that make some of the films useless and the low numbers of good film impressions (two to four), the film displaying the fewest artifacts, and with contact areas and pressure distributions consistent with apparently good trials, was analyzed for contact mechanics data.

Similar pressure measurements were also made using a peizo-resistive electronic pressure sensor. An electronic sensor specifically designed for the wrist joint was used (Model #4201, Tekscan Inc., South Boston, MA). Though the sensors have lower resolution than the Fuji-scale film, they allow for real time dynamic measurement of pressures, which eliminates the insertion and removal artifacts. The Tekscan sensor was thicker (7 mils or 0.18 mm) but much more flexible than the Pressurex film. Calibration of the Tekscan sensors followed similar methods as with the Pressurex film, with 10 levels of known loads applied over a known area using the same materials testing machine. The raw values of the sensor cells were then related to the applied pressure. Because the sensor was designed for the entire wrist joint (radiocarpal and ulnocarpal) and our study was focused on the radiocarpal joints only, each sensor was trimmed approximately 2 mm on each side, and resealed. This allowed insertion of the sensor with minimum disruption of the volar ligaments and capsule. The experimental setup and loading conditions for the Tekscan sensor were identical to those used for the Pressurex film. Measurements with the Tekscan system were repeated for three trials. The first trial always displayed contact areas and pressure distributions consistent with subsequent trials and was analyzed for contact mechanics data due to creep effects in the sensor that tend to result in slightly lower pressures on subsequent trials.

Specimen Imaging and Modeling.

MRI scans in this study were performed on a 9.4 T high resolution MRI scanner (Unity INOVA Animal Systems, Varian Inc., Palo Alto, CA) using a gradient recall echo sequence. Images were acquired from the frontal plane with a field of view of 60 mm × 120 mm, a data array of 512 × 1024 pixels, and a slice thickness of 1 mm. The number of slices was 30. The total scan time was approximately 24 min when the sequence parameters were set to relaxation time (TR) 800 ms, excitation time (TE) 7.81 ms, flip 45°, and number of excitations (NEX) 4. For each specimen, two image sets were obtained—“relaxed” and “grasp.” Relaxed specifically refers to the state of the wrist when no loading of the tendons or wrist is externally applied. Grasp (simulated) refers to when all five tendons were loaded, as described above.

Segmentation of the bones including their cartilage surfaces was performed on the relaxed image set (assumed undeformed), to obtain 3D surface models of the scaphoid, lunate, and radius bones including their cartilage. These surface models were then analyzed in the Joint_Model program [26].

Kinematics for the displacement-driven contact models were determined by image registration. From both image sets, each bone (without cartilage) was isolated on a black background (Fig. 2), resulting in six new image sets (one for each bone/configuration). The radius was chosen as the fixed/reference bone, and was used to align the coordinate systems of the two image sets by registration of the loaded radius to the unloaded radius. Registration was performed using 3D voxel matching of normalized mutual image intensity using Analyze 5.0 (AnalyzeDirect, Overland Park, KS). This algorithm sought to maximize the 3D correspondence between the image sets where they coincided, while ignoring areas that were not mutually shared in the images. The software provided a kinematic transformation that aligned the “match” image set to the “base” image set. The isolated grasp image sets with the loaded scaphoid and lunate were then transformed to the unloaded image coordinate system using the transformation from this initial registration. Lastly, the isolated relaxed image sets with the scaphoid were registered to the transformed, isolated grasp image set of the scaphoid. Similarly, the isolated relaxed image sets with the lunate were registered to the transformed, isolated grasp image set of the lunate. These final registrations provided the kinematic transformations to be applied in the modeling software as translation vectors and attitude vectors for rotation [30]. The tranformations placed the bone models (constructed from the relaxed image set) in the simulated grasp configuration.

Fig. 2.

Illustration of the bone isolation process prior to image registration. The isolated radius bone (right) was derived from the initial MR image (left), as would also be the case for the scaphoid and lunate.

Following determination of the kinematics, the contact mechanics were analyzed in the Joint_Model software. Contact was defined as interpenetration of the cartilage surfaces for the bones. Thus, the contact location and contact area were completely determined by the model geometry and the loaded configuration of the bones. The contact analysis proceeded with a linear contact rule, assuming (due to the relatively long time for the loading during the MRI scans) the cartilage was at a near-equilibrium state. Thus, we chose an effective compressive modulus of 4 MPa for the cartilage of all bones [26]. Cartilage thickness for each was assumed uniform at the average of the apparent thickness from our images, which we approximated at 1 mm. While these material parameters were simplifications, they provided an efficient and reasonable first order estimate of the contact parameters. For this material model, the local contact pressure was linearly related to the local interpenetration of the model (cartilage) surfaces, and pressure distributions could be computed over the contact area. Contact force was calculated by integrating the contact pressure over the contact area. In addition, average contact pressure was calculated as the contact force divided by contact area, and peak pressure was also evaluated. These contact parameters were compared to the experimental measures.

Direct Contact Area Measurement Using MRI Data.

The contact area was also compared to direct measurements of contact area from the MR images, which served as an additional verification of model contact area only. Note that this approach could also be used for verification of in vivo models. Segmentation of the grasp image set was performed only along the contact curve between the two adjacent bones being studied. The length of the curve was calculated and multiplied by the image slice thickness to obtain the effective contact area from each image. The contact areas for a given contact pair from all images were summed to determine the total contact area for the contact pair.

Validation Criteria.

Once data for contact force, contact area, average contact pressure, and peak pressure were collected from the Tekscan sensor, the Pressurex film, and the MRI-based models, they were compared qualitatively and quantitatively.

The validation criterion was met when the model data fell within two times the accuracy of the experimental measures [28]. Accuracy of the experimental contact measures from Pressurex film and the Tekscan sensor have been previously evaluated, but not fully characterized. Except in idealized testing, only the contact force can be precisely and externally measured. For the Tekscan sensors, the accuracy also depended on the specific sensor used, particularly the sensor resolution. Prior studies indicated errors from a low of 4% for Tekscan sensors to a high of 50% for Pressurex film for contact forces [31–33]. Based on these reported accuracy levels and our own internal accuracy study with six trials of idealized loading to evaluate force, we estimated average error in the radiocarpal joint of about 12%. Using this error level, we set the validation criteria for specimen-specific model outcome data within two times the expected experimental error (within 25%). It should be noted that determining accuracy and variability of local or peak pressure measurements is essentially intractable. The criterion of 25% was based primarily on measures of experimental variability of force data, and accuracy of peak pressure was likely lower, given the difficulty of measuring peak pressure. Thus, the criterion for peak pressure was likely more stringent than for other measures.

Results

Kinematics.

Kinematics for each specimen were unique, partially because of the unique geometry of each specimen’s bones, but primarily due to the variability of the bones in the unloaded state (initial separation). Carpal bone translations relative to the radius were on the order of 1–5 mm, and their rotations were all under 0.25°.

Contact Mechanics.

The MRI-based models for all four specimens predicted contact locations, contact areas, contact forces, and peak pressures similar to those measured in the experiments (Table 1 and Fig. 3). The relative size, shape, and location of radioscaphoid and radiolunate contact regions, their location, and the intensity of the pressure distribution was generally consistent between all measures, though Pressurex film appears to be laden with artifacts. The correspondence between Tekscan and model data was quite good. Both measures consistently placed the scaphoid and lunate contact areas at or near the dorsal rim of the radius (Fig. 3). The scaphoid tended to have higher contact pressure (except for Specimen 2) and a larger contact area than the lunate. It is important to remember that the experimental conditions (severed joint capsule, estimated tendon lines of actions and force magnitudes, etc.) did not necessarily replicate physiological grasp. As noted in the methods, the important thing in this study was that the loading conditions were repeatable between trials and for each different measurement technique, so that the results could be directly compared.

Table 1.

Comparison of the quantitative results from both experimental measures—Pressurex film (Film) and the Tekscan sensor (Tek)—and the MRI-based contact models (Model) for the radioscaphoid (RS) and radiolunate (RL) contact areas; peak contact pressures and contact forces

		Peak pressure (MPa)			Contact force (N)			Contact area (mm2)
Contact	Specimen	Film	Tek	Model	Film	Tek	Model	Film	Tek	Model	Direct
RS	S1	2.5	1.8	2.3 a , b	87	43	32 b	32.0	91.0	40.5 c	50.12
	S2	2.1	1.4	1.8 a	67	54	47 b	32.0	29.0	61.6	45.21
	S3	1.9	1.5	1.8 b	87	60	38	47.2	65.3	43.1 a	65.65
S4	1.2	2.8	2.8 b	33	54	69 b	32.8	70.1	61.2 b , c	77.50
RL	S1	3.8	1.0	2.1	75	35	34 b	20.0	76.0	31.3 c	31.14
	S2	0.9	1.7	2.4	29	36	37 b	13.8	18.1	31.4 c	29.20
	S3	0.6	1.4	1.4 b	11	39	37 b	18.7	51.7	52.8 b , c	50.94
S4	1.2	1.2	1.6	13	42	48 b	12.5	59.0	53.7 b , c	51.57

^aWithin 25% of the specimen-specific Pressurex film data.

^bWithin 25% of the specimen-specific Tekscan data.

^cWithin 25% of the specimen-specific direct contact area data.

Fig. 3.

Example comparison of pressure distribution from Pressurex film (a), (d), the Tekscan pressure sensor (b), (e), and the MRI-based models (c), (f) for Specimens 3 (a), (b), (c) and 4 (d), (e), (f). All images are oriented with dorsal at top, lunate fossa to the right, and scaphoid fossa to the left. Note the consistent dorsal contact measured by the Tekscan sensor and the MRI-based models.

Quantitatively, we first considered contact forces, because these measures can also be compared to the applied forces to simulate grasp. The sum of the contact forces was generally somewhat less than the 110 N applied across the wrist during simulated grasp, as expected due to partial load transfer between the ulna and the carpus (Fig. 4). The model data was consistently closer to the Tekscan experimental data, and total radiocarpal force met the validation criterion for all specimens with respect to the Tekscan measured contact forces. Pressurex force was found to be substantially higher than the applied force for Specimen 1 and substantially low for Specimen 4.

Fig. 4.

The sum of radiolunate and radioscaphoid joint contact forces across the wrist by specimen (S1–S4) for Pressurex film (Film), the Tekscan pressure sensor (Tek), and the MRI-based contact models (Model). The asterisks and braces indicate which model measurements met the validation criterion for total joint force and with respect to which experimental measures.

For individual articulation contact force, 7 of 8 articulations met the validation criterion with respect to Tekscan data (Table 1). For most specimens, the model predicted higher contact force in the radioscaphoid fossa compared to the radiolunate fossa. The average model-predicted force ratio between the two joints was 1.20 (Table 1). That value compared well with the average Tekscan ratio at 1.38, but the Pressurex ratio was much higher at 2.15. No model forces were within 25% of the Pressurex contact forces.

Like contact forces, average contact pressures from the model were generally closer to the Tekscan measures (Table 2). Only three of eight articulations met the validation criterion for average contact pressure with respect to Tekscan data and with respect to Pressurex data.

Table 2.

Comparison of the quantitative results from experimental measures of average contact pressure with the MRI-based contact models (Model) for the radioscaphoid (RS) and radiolunate (RL) contact areas, peak contact pressures and contact forces.

		Average contact pressure (MPa)
Contact	Specimen	Film	Tek	Model
RS	S1	2.7	0.5	0.8
	S2	2.1	1.9	0.8
	S3	1.8	0.9	0.9 b
S4	1.0	0.8	1.1 a
RL	S1	3.8	0.5	1.1
	S2	2.1	2.0	1.2
	S3	0.6	0.8	0.7 a , b
S4	1.0	0.7	0.9 a , b

^aWithin 25% of the specimen-specific Pressurex film data.

^bWithin 25% of the specimen-specific Tekscan data.

The peak pressure model data was somewhat less predictable—often falling between values from Pressurex film and the Tekscan sensor, but also sometimes higher or lower (Table 1). Still, the values for the experimental measures fell within a relatively narrow range (0.6–3.8 MPa), and the model predicted an even more narrow range (1.3–2.7 MPa). None but the very highest value, measured by Pressurex film, exceeded 2.7 MPa. Peak pressure model data met the validation criterion for only four of eight articulations with respect to Tekscan data and only two of eight with respect to Pressurex film data. Clearly, peak pressure was the most difficult parameter to measure, both experimentally and computationally.

Model contact area was compared to direct measurements from the grasp images, as well as to the Pressurex and Tekscan measures. Results between the contact model and the direct measurement agreed with remarkable precision for all radiolunate contact areas, with the largest difference being only 8% (Table 1, Fig. 5). Overall, contact area met the validation criterion for six of eight articulations with respect to direct image measurements, for three of eight articulations with respect to Tekscan data, and for one of eight articulations for Pressurex film data. The model contact areas generally corresponded with the Tekscan data better than Pressurex film data (Fig. 5). The Pressurex film contact areas were nearly always lower than all other measures, likely due to the measurement threshold, the difficulty in determining contact boundaries, and difficulties distinguishing true contact impressions from artifacts.

Fig. 5.

Contact area for all measurement methods compared by specimen (S1–S4), and by articulation (RS = radioscaphoid, RL = radiolunate). Model data matches the direct measurements well in all cases, and the Tekscan contact area is also consistently similar (except for S1).

Discussion

It was the goal of this study to validate the overall MRI-based modeling approach compared with experimental measurements. Most of model contact mechanics data for force and contact area met the validation criteria. That seven of eight individual articulation contact forces and four of four total radiocarpal joint forces from the model met the validation criterion (compared to Tekscan) indicated that the model results have similar force accuracy as the Tekscan sensor. Similarly, that six of eight model contact area measurements met the validation criterion with respect to contact area measured directly from grasp images indicated the reliability of model contact area measurement. These results demonstrated that MRI-based modeling can provide some acceptably accurate measures of contact mechanics. Data for peak and average contact pressures were more variable, and less than half of the contact pressure measurements met the validation criteria.

Contact location was qualitatively consistent between the model and the Tekscan data. Both measures consistently placed the scaphoid and lunate contact areas at or near the dorsal rim of the radius (Fig. 3). Thus, the data were internally consistent and these contact locations were consistent with prior in vivo and ex vivo measurements, as well [25,27].

For most specimens, the model predicted higher contact force in the radioscaphoid fossa as compared to the radiolunate fossa. The average model-measured force ratio between the two joints was 1.20 (Table 1), similar to our average Tekscan ratio at 1.38 and the ratio of 1.4 found in published work by others [34,35]. It has been previously shown that the percentage of total force carried in the scaphoid fossa of the wrist during grasp is about 50% of total applied force, and the total force carried through the lunar fossa during grasp is about 35% of total (the remaining 15% being carried through the ulnocarpal joint) [36,37]. Our data generally agreed with these percentages, especially the model data and Tekscan data. The model for specimen 4, however, had the sum of contact forces slightly above the applied loads, indicative of some modeling error (Fig. 4).

The fact that the peak pressure data from the model was somewhat less predictable than force or contact area should be tempered by the fact that peak pressure values still generally fell between the values from Pressurex film and the Tekscan sensor. Also, that the peak pressure values from the model fell within the range of 1.3–2.7 MPa, a range smaller than the values from the experimental measures (0.6–3.8 MPa), indicated that values of peak pressure from the model were likely reasonable. The model peak pressure data were at or under 2.7 MPa, which was similar to peak pressures previously measured in cadaveric experiments [34,35]. Even with the difficulty of measuring peak pressure, Specimen 3 was validated for peak pressure in both articulations. Overall, these findings provide evidence that the MRI-based modeling approach may be as accurate as the experimental measures of peak pressure.

This study had a number of limitations. One limitation to achieving the validation goal was the relatively high level of variability associated with the experimental measures. In addition, the assumption of uniform cartilage thickness affected estimates (plural, estimates) of the local contact pressures. Particularly this may have affected the accuracy of peak contact pressure measurements. In addition, because of the variability of the initial conditions and the dissection needed to measure pressure in the cadaveric experiment, no specific conclusions could be drawn from the kinematic data.

The contact area data from the Tekscan sensor may have been somewhat low due to the short length of time for loading prior to measurement. For the Tekscan sensor in particular, the contact measurements were made as soon as the load was applied and the wrist was stable, to avoid potential creep in the measurement. The MRI models were based on prolonged loading over the course of the 24 min scan. Thus, different amounts of cartilage relaxation could have been a factor in differences between model data and experimental data.

Further, with regard to contact area measurements, the Tekscan sensor for the wrist has a relatively low resolution, which could contribute to inaccuracy. Pressurex film measurements consistently underestimated contact area with respect to all other measures, which may have been partly due to the fact that pressures under 0.5 MPa will not register on the film. These are clear examples of limitations of both of the measurement systems in the context of this study.

While the majority of validation studies in the literature have compared the model results to data from the literature or compared data from a single specimen-specific model to experimental measures, we have performed specimen-specific validation analyses for four specimens. The variability of the results between specimens and the various measures to be validated illustrates the need for thorough and careful validation studies when considering a single specimen. This study also illustrates the difficulty in quantitatively validating multiple models for multiple outcome measures.

The results of this work are promising. The specimens behaved consistently, and the MRI-based models yielded reasonable results, showing that an MRI-based model for calculating in vivo joint contact mechanics is feasible. The specimens tested showed consistent results, and the MRI-based model data predicted results sufficiently similar to the experimental results to consider the model accurate for the prediction of contact area and contact force. The average contact pressure and, especially, the peak contact pressure are the most difficult outcome measures to obtain accurately (by any means), but for prediction of OA risk, peak pressure is perhaps the most important outcome measure. The variability within and between experimental measures indicates that peak pressure is also experimentally the least accurate. While not strictly validated, average and peak contact pressures from the MRI-based models appear to have similar accuracy to current experimental measures. Overall, the MRI-based modeling technique appears to be as accurate as the experimental measures it was compared to for this validation study, and it shows excellent promise for future in vivo studies.