Evaluation of a Validation Method for MR Imaging-Based Motion Tracking Using Image Simulation
© Kevin M. Moerman et al. 2010
Received: 1 May 2009
Accepted: 20 July 2009
Published: 13 September 2009
Magnetic Resonance (MR) imaging-based motion and deformation tracking techniques combined with finite element (FE) analysis are a powerful method for soft tissue constitutive model parameter identification. However, deriving deformation data from MR images is complex and generally requires validation. In this paper a validation method is presented based on a silicone gel phantom containing contrasting spherical markers. Tracking of these markers provides a direct measure of deformation. Validation of in vivo medical imaging techniques is often challenging due to the lack of appropriate reference data and the validation method may lack an appropriate reference. This paper evaluates a validation method using simulated MR image data. This provided an appropriate reference and allowed different error sources to be studied independently and allowed evaluation of the method for various signal-to-noise ratios (SNRs). The geometric bias error was between 0– voxels while the noisy magnitude MR image simulations demonstrated errors under 0.1161 voxels (SNR: 5–35).
The body responds to mechanical loading on several timescales (e.g., [1, 2]), but in vivo measurement of critical parameters such as muscle load, joint reaction force, and tissue stress/strain is usually not possible [3, 4]. In contrast, suitably validated computational models can predict all of these parameters, and they are therefore a powerful tool for understanding the musculoskeletal system [4, 5] and are in use in diverse applications from impact biomechanics [6, 7] to rehabilitation engineering [8, 9], surgical simulation [10, 11], and soft tissue drug transport .
Skeletal muscle tissue in compression is nonlinear elastic, anisotropic, and viscoelastic, and a constitutive model with very good predictive capabilities for in vitro porcine muscle has been proposed [13, 14]. However, validating this model for living human tissue presents significant difficulties. Some authors have used indentation tests on skeletal muscle [15, 16], but the tissue was then assumed to be isotropic and linear in elastic and viscoelastic properties. In contrast, noninvasive imaging methods that allow detailed measurement of human soft tissue motion and deformation (due to known loading conditions) combined with inverse finite element (FE) analysis allow for the evaluation of more complex constitutive models.
The work presented here is part of a study aiming to use indentation tests on the human arm, tagged Magnetic Resonance (MR) imaging and inverse FE analysis to determine the mechanical properties of passive living human skeletal muscle tissue using the constitutive model described in [13, 14].
Recently the potential of using surface deformation measurements from 3D digital image correlation to assess mechanical states throughout the bulk of a tissue has been shown . However MR imaging combined with deformation tracking techniques can provide 3D deformation data throughout the tissue volume and is ideal for the evaluation of constitutive models such as [13, 14]. MR imaging has been used to study skin , heart , and recently also rat skeletal muscle  (though a simplified Neo-Hookean model was applied).
The techniques for tracking tissue deformation from (e.g., tagged) MR imaging are complex and require validation using an independent measure of deformation. Since physically implanting markers is not feasible and anatomic landmarks are either absent or difficult to track, alternative methods have been employed. Young et al.  recorded angular displacement of a silicone gel phantom using tagged MR images and evaluated the results using FE modelling and 2D surface deformation derived from optical tracking of lines painted on the phantom surface. Similarly, Moore et al.  used optical tracking of surface lines on a silicone rubber phantom to validate MR-based deformation measures. However simple tensile stretch was applied and only a 2D measure of surface deformation was used. There were also temporal synchronisation issues between the optical and MR image data. In both of the optical validations studies above the error related to the optical tracking method was not quantified. Other authors have used implantable markers. For instance Yeon et al.  used implanted crystals and sonomicrometric measurements for validation of tagged MR imaging of the canine heart. However the locations of the crystals were verified manually by mapping with respect to surface cardiac landmarks in the excised heart and matching problems between MR and sonomicrometric measurements occurred. Neu et al. [24, 25] evaluated a tagged MR imaging-based deformation tracking technique for cartilage using spherical marker tracking in a silicone soft tissue phantom. However the marker centres were determined by manually fitting a circle to each marker in two orthogonal directions and imaging was performed on excised tissue samples at high resolution (over 32 voxels across marker diameter) using a nonclinical 7.05T scanner.
This paper shows that validation of in vivo medical imaging techniques and image processing algorithms is challenging partially due to the lack of appropriate reference data. Although experimental validation methods using soft tissue MR imaging phantoms can be developed, the data derived from these often suffers uncertainties similar to those present in the target soft tissue. Therefore the validation method itself often lacks an appropriate reference. In this paper a novel technique for the validation of a 3D MR imaging-based motion and deformation tracking technique, applicable to 3D deformation, is presented. The validation method, based on marker tracking, was evaluated (and validated) using simulated magnitude MR image data because this allows full control and knowledge of marker locations and thus provides the final real "gold standard.'' It addition this allows for the independent analysis of geometric bias and of method performance across a wide range of realistic noise conditions.
2.1. The Tissue Phantom
2.2. MR Imaging
The type of image data used in the current study is T2 magnitude MR images. Deformation can be measured using marker tracking methods applied to full volume scans taken at each deformation step. A full volume scan was performed on the tissue phantom using a 3T scanner (Philips Achieva 3T, Best, The Netherlands). Cubic 0.5 mm voxels were used and the data was stored using the Digital Imaging and Communication in Medicine (DICOM) format. Figures 1(a) and 1(b) show an example of an MR image and tagged MR image of a region of the phantom. The voxel intensities of the images are 9 bit unsigned integers with values ranging from 0 to 511. The data was imported into Matlab 7.4 R2007a (The Mathworks Inc., USA) for image processing. The image data was normalised producing an average gel intensity of 0.39, while the marker intensity was zero.
2.3. Marker Tracking Method
To track the movement of markers from the 3D MR data an image processing algorithm was developed in Matlab (The Mathworks Inc., USA). The centre point of each marker at each time step can be found using 3 main steps: masking, adjacency grouping, and centre point calculation.
The ACG contains all the directly adjacent voxels of the voxel (its direct upper, lower, front, back, left, and right neighbours). Any voxel with coordinates is added to the marker group when its intensity is lower than and its coordinates are found within one of the ACGs of the marker . When a voxel is added to the marker group its ACG is added to the set of ACGs belonging to and this process is repeated. Voxels are added to a marker group until the group is no longer growing.
Figure 2(b) shows how, starting with one central voxel, the surrounding low intensity voxels within the coordinate group are added and when this is repeated all voxels representing the marker are grouped. After grouping, the dimensions and number of voxels of the object were compared to what is expected for normal markers (e.g., a diameter of under 6 voxels and consisting of under 250 voxels) to filter out possible objects other than markers.
Here represents a threshold which for a noiseless image could be set equal to the gel intensity (the weight then represents the volume fraction of marker material present in the voxel). The condition is added that when is larger than the weight .
2.4. Evaluation of Marker Tracking Method Using Simulated Magnitude MR Image Data
The marker tracking method was evaluated using simulated magnitude MR image data because this allows full control and knowledge of marker locations and thus provides the final real "gold standard.'' The simulated data also allow one to isolate and study errors from different sources. The marker tracking method was evaluated using algorithms developed in Matlab (The Mathworks Inc., USA) and involves the following steps: simulation of a noiseless image and analysis of geometric bias, and simulation of noisy magnitude MR data and analysis of the noise effects. The final noisy image data allows one to evaluate the performance of the method under varying noise conditions while the noiseless image allows for evaluation of the geometric bias implicit in the method.
The geometric bias was investigated by simulating markers with their centre points coinciding with various locations within an OCV in the absence of noise. Due to the symmetry in the appearances as discussed above, simulations were performed in 1 octant of the OCV only using a grid of points. For visualisation purposes the results were then mirrored to obtain bias measures across the full OCV (similar to Figure 5(a)) producing a grid. A finer grid was then applied around the maximum bias to closely approximate the location of the real maximum bias. This process was repeated until the found maximum no longer varied significantly.
The results are presented in two steps: evaluation of the geometric bias in the marker tracking method, and evaluation of the performance on the marker tracking method in the presence of noise.
Several MR imaging-based motion tracking algorithms have been proposed in the literature, for example, tagged MR imaging  and phase contrast MR imaging , but these all rely upon validation of the algorithms proposed. A review of the literature showed that the validation methods used for existing techniques are frequently incomplete, and this paper presents a novel validation method for MR imaging based on motion tracking using a marker tracking algorithm which itself is validated against simulated MR image data. Simulated data was generated for the noise-free case as well as for a variety of different Rician distributed noise levels. The noise-free image data allowed analysis of the error related to the geometric bias independently from other error sources.
Therefore the method proved to be robust with geometric bias errors of between voxels and errors due to noise remaining below 0.1127 voxels for all cases simulated with signal-to-noise ratios from 5 to 35. These results were achieved for a global threshold value . However altering the threshold value based on the SNR may result in a significant increase in accuracy. The optimum value for a certain SNR can be determined using MR data simulations. Using simulations the error can be minimised for a given SNR by adjusting the value.
The method proposed in this paper has two main advantages. The first is that the data used for validation is simulated and therefore can be chosen to have desired levels of noise. This permitted evaluation of the marker tracking method for different levels of noise which has not been done previously. Secondly, since this validation method is based on MR imaging, the marker tracking experiment and the MR imaging-based motion and deformation tracking can all be performed at the same time within the MR scanner.
Although this method has been developed for application to tagged MR imaging on the upper arm, the methods presented here are not limited to this application and can be applied to validate other types of MR imaging-based motion and deformation tracking techniques. Furthermore, these methods are independent of the chosen phantom shape.
A novel marker tracking method has been presented and validated using simulated MR image data. The marker tracking method is robust and the maximum geometric bias was voxels while the error due to noise remains below 0.1127 voxels for Rician noise distributions with signal-to-noise ratios from 5 up to 35. This appears to be the only marker tracking algorithm suitable for the validation of MR-based motion and deformation tracking of soft tissue which has been validated against a "gold standard.''
This work was funded by a Research Frontiers Grant (06/RF/ENMO76) awarded by Science Foundation Ireland.
- Lissner H, Lebow M, Evans F: Experimental studies on the relation between acceleration and intracranial pressure changes in man. Surgery Gynecology & Obstetrics 1960, 111: 329-338.Google Scholar
- McNamara LM, Prendergast PJ: Bone remodelling algorithms incorporating both strain and microdamage stimuli. Journal of Biomechanics 2007, 40(6):1381-1391. 10.1016/j.jbiomech.2006.05.007View ArticleGoogle Scholar
- Davis J, Kaufman KR, Lieber RL: Correlation between active and passive isometric force and intramuscular pressure in the isolated rabbit tibialis anterior muscle. Journal of Biomechanics 2003, 36(4):505-512. 10.1016/S0021-9290(02)00430-XView ArticleGoogle Scholar
- Erdemir A, McLean S, Herzog W, van den Bogert AJ: Model-based estimation of muscle forces exerted during movements. Clinical Biomechanics 2007, 22(2):131-154. 10.1016/j.clinbiomech.2006.09.005View ArticleGoogle Scholar
- Marjoux D, Baumgartner D, Deck C, Willinger R: Head injury prediction capability of the HIC, HIP, SIMon and ULP criteria. Accident Analysis & Prevention 2008, 40(3):1135-1148. 10.1016/j.aap.2007.12.006View ArticleGoogle Scholar
- Muggenthaler H, von Merten K, Peldschus S, et al.: Experimental tests for the validation of active numerical human models. Forensic Science International 2008, 177(2-3):184-191. 10.1016/j.forsciint.2007.12.005View ArticleGoogle Scholar
- Ivancic PC, Ito S, Panjabi MM: Dynamic sagittal flexibility coefficients of the human cervical spine. Accident Analysis & Prevention 2007, 39(4):688-695. 10.1016/j.aap.2006.10.015View ArticleGoogle Scholar
- Linder-Ganz E, Shabshin N, Itzchak Y, Gefen A: Assessment of mechanical conditions in sub-dermal tissues during sitting: a combined experimental-MRI and finite element approach. Journal of Biomechanics 2007, 40(7):1443-1454. 10.1016/j.jbiomech.2006.06.020View ArticleGoogle Scholar
- Linder-Ganz E, Shabshin N, Itzchak Y, Yizhar Z, Siev-Ner I, Gefen A: Strains and stresses in sub-dermal tissues of the buttocks are greater in paraplegics than in healthy during sitting. Journal of Biomechanics 2008, 41(3):567-580. 10.1016/j.jbiomech.2007.10.011View ArticleGoogle Scholar
- Lim Y-J, De S: Real time simulation of nonlinear tissue response in virtual surgery using the point collocation-based method of finite spheres. Computer Methods in Applied Mechanics and Engineering 2007, 196(31-32):3011-3024. 10.1016/j.cma.2006.05.015View ArticleMATHGoogle Scholar
- Audette MA, Hayward V, Astley O, Doyon M, McCallister GA, Chinzei K: A PC-based system architecture for real-time finite element-based tool-specific surgical simulation. Proceedings of the 18th International Congress and Exhibition on Computer Assisted Radiology and Surgery (CARS '04), June 2004, International Congress Series 1268: 378-383.Google Scholar
- Wu PI, Edelman ER: Structural biomechanics modulate intramuscular distribution of locally delivered drugs. Journal of Biomechanics 2008, 41(13):2884-2891. 10.1016/j.jbiomech.2008.06.025View ArticleGoogle Scholar
- Van Loocke M, Lyons CG, Simms CK: A validated model of passive muscle in compression. Journal of Biomechanics 2006, 39(16):2999-3009. 10.1016/j.jbiomech.2005.10.016View ArticleGoogle Scholar
- Van Loocke M, Lyons CG, Simms CK: Viscoelastic properties of passive skeletal muscle in compression: stress-relaxation behaviour and constitutive modelling. Journal of Biomechanics 2008, 41(7):1555-1566. 10.1016/j.jbiomech.2008.02.007View ArticleGoogle Scholar
- Gefen A, Gefen N, Linder-Ganz E, Margulies SS: In vivo muscle stiffening under bone compression promotes deep pressure sores. Journal of Biomechanical Engineering 2005, 127(3):512-524. 10.1115/1.1894386View ArticleGoogle Scholar
- Palevski A, Glaich I, Portnoy S, Linder-Ganz E, Gefen A: Stress relaxation of porcine gluteus muscle subjected to sudden transverse deformation as related to pressure sore modeling. Journal of Biomechanical Engineering 2006, 128(5):782-787. 10.1115/1.2264395View ArticleGoogle Scholar
- Moerman KM, Holt CA, Evans SL, Simms CK: Digital image correlation and finite element modelling as a method to determine mechanical properties of human soft tissue in vivo. Journal of Biomechanics 2009, 42(8):1150-1153. 10.1016/j.jbiomech.2009.02.016View ArticleGoogle Scholar
- Tada M, Nagai N, Yoshida H, Maeno T: Iterative FE analysis for non-invaseice material modeling of a fingertip with layered structure. Proceedings of the Eurohaptics, 2006, Paris, FranceGoogle Scholar
- Walker JC, Ratcliffe MB, Zhang P, et al.: Magnetic resonance imaging-based finite element stress analysis after linear repair of left ventricular aneurysm. Journal of Thoracic and Cardiovascular Surgery 2008, 135(5):1094-1102. 10.1016/j.jtcvs.2007.11.038View ArticleGoogle Scholar
- Ceelen KK, Stekelenburg A, Mulders JLJ, et al.: Validation of a numerical model of skeletal muscle compression with MR tagging: a contribution to pressure ulcer research. Journal of Biomechanical Engineering 2008, 130(6):-8.View ArticleGoogle Scholar
- Young AA, Axel L, Dougherty L, Bogen DK, Parenteau CS: Validation of tagging with MR imaging to estimate material deformation. Radiology 1993, 188(1):101-108.View ArticleGoogle Scholar
- Moore CC, Reeder SB, McVeigh ER: Tagged MR imaging in a deforming phantom: photographic validation. Radiology 1994, 190(3):765-769.View ArticleGoogle Scholar
- Yeon SB, Reichek N, Tallant BA, et al.: Validation of in vivo myocardial strain measurement by magnetic resonance tagging with sonomicrometry. Journal of the American College of Cardiology 2001, 38(2):555-561. 10.1016/S0735-1097(01)01397-3View ArticleGoogle Scholar
- Neu CP, Hull ML, Walton JH: Error optimization of a three-dimensional magnetic resonance imaging tagging-based cartilage deformation technique. Magnetic Resonance in Medicine 2005, 54(5):1290-1294. 10.1002/mrm.20669View ArticleGoogle Scholar
- Neu CP, Hull ML, Walton JH, Buonocore MH: Toward an MRI-based method to determine three-dimensional deformations in articular cartilage. Proceedings of the Summer Bioengineering Conference, 2003, Key Biscayne, Fla, USAGoogle Scholar
- Goldstein DC, Kundel HL, Daube-Witherspoon ME, Thibault LE, Goldstein EJ: A silicone gel phantom suitable for multimodality imaging. Investigative Radiology 1987, 22(2):153-157. 10.1097/00004424-198702000-00013View ArticleGoogle Scholar
- Augenstein KF, Cowan BR, LeGrice IJ, Nielsen PMF, Young AA: Method and apparatus for soft tissue material parameter estimation using tissue tagged magnetic resonance imaging. Journal of Biomechanical Engineering 2005, 127(1):148-157. 10.1115/1.1835360View ArticleGoogle Scholar
- Denney TS Jr., Prince JL, Lopez MJ, McVeigh ER: Optimal tag pattern validation using magnetic resonance imaging. Proceedings of the IEEE International Conference on Image Processing (ICIP '94), 1994, Austin, Tex, USAGoogle Scholar
- Fahmy AS, Krieger A, Osman NF: An integrated system for real-time detection of stiff masses with a single compression. IEEE Transactions on Biomedical Engineering 2006, 53(7):1286-1293. 10.1109/TBME.2006.873690View ArticleGoogle Scholar
- Kraitchman DL, Young AA, Chang C-N, Axel L: Semi-automatic tracking of myocardial motion in MR tagged images. IEEE Transactions on Medical Imaging 1995, 14(3):422-433. 10.1109/42.414606View ArticleGoogle Scholar
- Osman NF: Detecting stiff masses using strain-encoded (SENC) imaging. Magnetic Resonance in Medicine 2003, 49(3):605-608. 10.1002/mrm.10376View ArticleGoogle Scholar
- Sampath S, Parthasarathy V, Prince JL: A phantom validation of the FastHARP pulse sequence. Proceedings of the IEEE International Symposium on Biomedical Imaging, 2002Google Scholar
- Wyman BT: Mechanical Evaluation of the Paced Heart Using Tagged Magnetic Resonance Imaging. The Department of Biomedical Engineering, The Johns Hopkins University, Baltimore, Md, USA; 1999.Google Scholar
- Young AA, Kraitchman DL, Dougherty L, Axel L: Tracking and finite element analysis of stripe deformation in magnetic resonance tagging. IEEE Transactions on Medical Imaging 1995, 14(3):413-421. 10.1109/42.414605View ArticleGoogle Scholar
- Henkelman RM: Measurement of signal intensities in the presence of noise in MR images. Medical Physics 1985, 12(2):232-233. 10.1118/1.595711View ArticleGoogle Scholar
- Gudbjartsson H, Patz S: The Rician distribution of noisy MRI data. Magnetic Resonance in Medicine 1995, 34(6):910-914. 10.1002/mrm.1910340618View ArticleGoogle Scholar
- Cárdenas-Blanco A, Tejos C, Irarrazaval P, Cameron I: Noise in magnitude magnetic resonance images. Concepts in Magnetic Resonance A 2008, 32(6):409-416.View ArticleGoogle Scholar
- Aja-Fernandez S, Alberola-Lopez C, Westin C-F: Noise and signal estimation in magnitude MRI and Rician distributed images: a LMMSE approach. IEEE Transactions on Image Processing 2008, 17(8):1383-1398.MathSciNetView ArticleGoogle Scholar
- Rice SO: Mathematical analysis of random noise. Bell Systems Technical Journal 1944, 23: 282-332.MathSciNetView ArticleMATHGoogle Scholar
- Rice SO: Mathematical analysis of random noise. Bell Systems Technical Journal 1945, 24: 46-156.MathSciNetView ArticleMATHGoogle Scholar
- Axel L, Montillo A, Kim D: Tagged magnetic resonance imaging of the heart: a survey. Medical Image Analysis 2005, 9(4):376-393. 10.1016/j.media.2005.01.003View ArticleGoogle Scholar
- Gatehouse PD, Keegan J, Crowe LA, et al.: Applications of phase-contrast flow and velocity imaging in cardiovascular MRI. European Radiology 2005, 15(10):2172-2184. 10.1007/s00330-005-2829-3View ArticleGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.