- Email: [email protected]
In-vivo deformation measurements of the human heart by 3D Digital Image Correlation
Journal of Biomechanics 48 (2015) 2217–2220
Contents lists available at ScienceDirect
Journal of Biomechanics journal homepage: www.elsevier.com/locate/jbiomech www.JBiomech.com
Short communication
In-vivo deformation measurements of the human heart by 3D Digital Image Correlation Mikko Hokka a,n, Nikolas Mirow b, Horst Nagel c, Marc Irqsusi b, Sebastian Vogt b, Veli-Tapani Kuokkala a a
Tampere University of Technology, Department of Materials Science, Finland Heart Surgery, Universitätsklinikum Gießen und Marburg GmbH, Marburg, Germany c LaVision Ltd, Göttingen, Germany b
art ic l e i nf o
a b s t r a c t
Article history: Accepted 13 March 2015
Fast and accurate measurements of the kinetics and deformation of the heart during cardiac surgery can be useful for assessing the best strategies for the protection of the myocardium. While measurements based on ultrasonic technology such as the transesophageal echocardiography are rapidly developing in this direction, also other analysis methods based on optical imaging have been developed within the recent decade. The improved quality of digital cameras and increased computational power of personal computers have led to the development of deformation analysis method known as Digital Image Correlation (DIC). This paper presents preliminary results on the application of the DIC technique on analysing of the movement and deformation of the myocardial movement during a cardiopulmonary bypass surgery. The results show that the natural pattern of the heart should be sufficient for DIC, but better and more accurate results could be obtained with improved contrast conditions. DIC has a potential to be used as a sensitive tool for the surgeon to monitor the cardiac function. & 2015 Elsevier Ltd. All rights reserved.
Keywords: Digital Image Correlation Myocardium Deformation
1. Introduction Online monitoring of the movements and deformation of the ventricles and atria of the human heart may be useful when assessing different myocardial protection strategies during cardiac surgery. Recently, various imaging methods based on ultrasonic measurements, such as transesophageal echocardiography and tissue Doppler techniques (Shimony et al., 2014; Pauliks et al., 2014) have been used for analysing the movement and deformation of the heart. In addition, other experimental and numerical methods based on magnetic resonance tissue tagging and finite element simulations are being used for estimating the deformation of the heart (Ibrahim, 2011; Dorri et al., 2006). Strain and strain rate measurements of the right ventricle (RV) using echocardiography have considerably enhanced the understanding of the RV function (La Gerche et al., 2010). During the operation, however, echocardiography is not easily applicable to imaging of the (RV) or the atria. The monitoring of the RV requires advanced skill from the investigator, and all investigations are performed by an invasive transesophageal probe that occasionally induces gastrointestinal injuries and bleeding. Clinical difficulties in assessing n
Corresponding author. Tel.: þ 358 408490132 E-mail address: mikko.hokka@tut.fi (M. Hokka).
http://dx.doi.org/10.1016/j.jbiomech.2015.03.015 0021-9290/& 2015 Elsevier Ltd. All rights reserved.
the RV function using echocardiography are discussed by Teske et al. (2009). RV dysfunction can cause severe complications and even death during weaning from extracorporeal circulation (ECC) and mechanical ventilation after the cardiosurgical operation (e.g. bypass grafting, valve replacement). This is especially the case for patients who in addition to the acquired heart decease are additionally suffering from pulmonary hypertension or chronic obstructive pulmonary disease (COPD) (Porhomayon et al. 2012). Therefore, there is a need for fast and accurate analysis of the local deformation of the heart that can be used for strain and strain rate measurements of the right side of the heart. Digital Image Correlation (DIC) has been used for solving of various engineering and materials science problems within the past decades, and recently also an increasing number of medical applications have been studied using DIC. Moerman et al. (2009) studied the properties and mechanical response of artificial soft tissue using a combined modelling and DIC approach. Han et al. (2012) used DIC on ultrasonic images for diagnosis of breast cancer, whereas Libertiaux et al. (2011) studied the compressive behaviour of brain tissue using DIC. Recently, Lionello et al. (2014) developed tools for generating speckle patterns for DIC measurements of soft tissues. In 3D DIC, the cameras are observing the target from different angles, and a stereo image is reconstructed from two or more individual images using (typically) a pinhole camera model. The
2218
M. Hokka et al. / Journal of Biomechanics 48 (2015) 2217–2220
pinhole model is typically calibrated using an image of a calibration target obtained with an identical optical setup that was used for the imaging of the analysed target. The pinhole model itself is used for mapping of the two or more camera coordinate systems and the world (absolute) coordinate system. The deformation analysis is based on tracking of the movement of an image subset between consecutive images obtained as a function of time. The movement of the image subset gives the 3D displacement vectors, which can be obtained on the surface of the target at a very high spatial resolution. The tracking algorithm uses cross-correlation to find the exact location of the subset in the next image, and the deformation of the subset is taken into account using a shape function that allows the image subset to deform as a function of time. The displacements of the subsets can be obtained at a precision less than one pixel by interpolation of the grey level values between the pixels. A good overview of the DIC method, more detailed mathematical descriptions of the calibration, pinhole camera model, cross correlation, shape functions etc. as well as several examples can be found in, for example, Ref. (Sutton et al., 2009).
2. Materials and methods
positioning of the cameras were done simply by estimating where the heart of the patient would most likely be during the operation. Due to the fact that the exact location of the area of interest was not known, 50 mm lenses were used in acquiring the images, so that a large area would be covered in the images. Intraoperative image collection was carried out at the rate of 10 frames per second using LaVision E-Lite 5MPix digital cameras. The calibration of the DIC system was done by obtaining several images of the 3D calibration plate with 2.2 mm dots, 10 mm distance between the dots, and 2 mm distance between the two planes. No extra pattern was added on the surface of the myocardium and the pattern matching was carried out on the natural pattern of the surface. The DAVIS software (DaVis 8.20, LaVision Ltd) was used for obtaining the images as well as for the displacement and strain calculations. The exact mathematical formulation used for pattern matching, displacement and strain calculations and region growth algorithms are presented elsewhere (Fleet and Weiss, 2006; Bouguet, 2000). The calibration of the DIC setup had to be done prior to and/or after the surgery. Therefore, the exact distance from the calibration plate to the final/actual target was not known at the time when the calibration images were obtained. This distance was estimated by first measuring the distance from the calibration target and later on from the patient's chest to the OR floor. This estimated distance was used for the first calibration, which was then iteratively improved by transforming the calculated surface to Z¼ 0 mm position. A reasonable calibration was obtained with a standard deviation of the fit of 0.46 pixels, where one pixel corresponds to a distance of 146.8 μm.
3. Results and discussion The displacement vectors were calculated from the masked area of interest by choosing several seed points where the correlation
The experiments were carried out at the Universitätsklinikum Marburg, Germany. The image correlation system was setup in the operating theatre approximately 45 min prior to the operation. The cameras were positioned above and in front of the patient's chest, as shown in Fig. 1. However, the cameras were positioned and focused prior to bringing the patient to the operating room (OR) and therefore the focusing and
Fig. 1. Positioning of the cameras in the operating room.
Fig. 2. Stereo reconstruction errors at 0.1 s and 0.6 s corresponding to (a) average and (b) maximum errors. The images corresponding to average and maximum cases were selected by visual comparison of the images.
M. Hokka et al. / Journal of Biomechanics 48 (2015) 2217–2220
2219
Fig. 3. (a–c) are the minimum (compression) principal (Lagrange) strains on the surface of the heart overlaid on the reconstructed 3D image, and (d–f) are the maximum (tension) principal strains. The images correspond to time points (a and d) of 0.2 s, (b and e) of 0.4 s, and (c and f) of 0.6 s.
algorithm starts constructing the stereo images. The whole visible area of the heart was masked and the seed points were manually selected by trial and error method. Due to the rather poor quality of the surface pattern, a very large subset of 121 pixels with a step size of 1 pixel was used for the pattern matching. Large subsets include more image data and therefore increase the accuracy of the displacement measurement. However, the large subset size and small step size also dramatically increase the computation time. Using such a small step size will not significantly increase the spatial resolution of the method, but it was found to stabilise the area growth algorithm when analysing low quality patterns. This significantly reduces the number of required seed points. Furthermore, the high intensity focused lights used in the OR (shown in Fig. 1) resulted in glare from the wet surface of the heart. Normally a more diffused light source is preferred when recording images for DIC. The glare changes location in consecutive images and effectively prevents image correlation in areas with strong glare. In the images obtained in this work, strong glare occurred in the top parts of the images or on the lower side of the heart. Due to the low contrast of the pattern and the glare issues, most of the calculations were carried out on the surface of the right atrium, and the right ventricle was analysed only partly. Despite the poor contrast of the surface pattern and the glare problems, the stereo reconstruction errors remained relatively small and larger errors were observed only at some “hot spots”, where the pattern matching algorithm did not find a good match in the images obtained with the two cameras. Fig. 2a and b shows the 3D reconstruction overlaid with the stereo reconstruction error in pixels obtained at 0.1 s and 0.6 s from the beginning of the recording. The stereo reconstruction errors in the obtained image pairs were analysed simply by visually comparing the images where the stereo reconstruction error is overlaid with the 3D image. On average, the errors obtained in all images are very close to that shown in Fig. 2a, except for the images with maximum deformation, which are shown in Fig. 2b. The maximum errors are found repeatedly at locations with maximum deformation. At this stage the stereo reconstruction
error remains small around the atria, and the maximum errors are found at the centre of the analysed region. In this region the stereo reconstruction errors are as high as 4 pixels, which is due to the poor quality and the large deformation of the pattern. The low quality contrast patterns in analysing with DIC are a common problem for soft tissues. Gao and Desai (2009) studied porcine liver tissues using DIC with natural patterns. They carried out translation tests to evaluate the quality of the natural pattern. The sample was simply translated without any deformation and the displacements of the surface were calculated using 2D global DIC. The standard deviation of displacements for one image pair in their test varied between 0.04 and 0.44 pixels, and the cumulative standard deviation for 15 images was 0.69 pixels. The stereo reconstruction errors in the current work compare well with the values of Gao et al. but only when the deformation of the heart is small. The simple way to improve the accuracy is to improve the contrast pattern. Several research groups have made considerable efforts in improving the contrast patterns of soft tissues. Miri et al. (2012) were able to measure the deformation of porcine vocal folds to strains up to 50%. However, they used commercial tissue dyes to improve contrast of the pattern. Other groups have improved the contrast of the pattern by powders (Myers et al., 2010) or powders mixed with gels (Thompson et al., 2007). However, the improved pattern for the in-vivo measurements of the human heart remains a challenge due to the requirements for a sterile biocompatible or removable pattern. This problem needs to be addressed in future research regarding DIC analysis of cardiac function during open heart surgery. Fig. 3 shows a short time series of the DIC strain maps overlaid with the reconstructed 3D image. The images a–c show the minimum principal strain, and the images d–f show the maximum principal strain. The images were obtained at 0.1 s intervals, but only after every second image is shown. On average, the minimum principal strain varies between 0% and 10%. Clearly the region between the right atrium and the right ventricle deforms strongly, but unfortunately the pattern quality does not allow for a detailed
2220
M. Hokka et al. / Journal of Biomechanics 48 (2015) 2217–2220
analysis of this region. The maximum principal strains range on average between 5% and 10% of tensile deformation. 4. Conclusions Deformation of the heart was obtained using Digital Image Correlation of the optical images acquired intraoperatively during cardiopulmonary bypass surgery. The natural surface pattern of the myocardium was sufficient for the calculation of the 3D displacement vectors and the strains on its surface. The natural pattern was, however, not sufficient for the construction of the 3D images for the whole area of interest, and therefore not all areas could be studied. For improving the analysis and reducing the calculation time, the contrast of the natural pattern needs to be improved either by better illumination conditions or by developing a sterile biocompatible speckle pattern. Also, other methods requiring less contrast patterns, such as feature tracking algorithms might provide more use as an online tool during the surgery. Conflict of interest None. Acknowledgements Dr. Sven Curtze (Nacon – nanoAnalysis Consulting Inc.) is acknowledged for his efforts to make these measurements possible and for the encouragement to undertake these experiments. References Bouguet, J.-Y., 2000. Pyramidal Implementation of the Lucas Kanade Feature Tracker Description of the Algorithm. Intel Corporation Microprocessor Research Labs, Available online at.
Dorri, F., Niederer, P.F., Lunkenheimer, P.P., 2006. A finite element model of the human left ventricular systole. Comput. Methods Biomech. Biomed. Eng. 9, 319–341. Fleet, D., Weiss, Y., 2006. Optical flow estimation. In: Nikos, Paragious, Yunmei, Chen, Olivier D, Faugeras, Paragios, et al. (Eds.), Handbook of Mathematical Models in Computer Vision. Springer, USA. Gao, Z., Desai, J., 2009. Estimating zero-strain states of very soft tissue under gravity loading using digital image correlation. Med. Image Anal. 14, 126–137. Han, Y., Kim, D.-W., Kwon, H., 2012. Application of digital image cross-correlation and smoothing function to the diagnosis of breast cancer. J. Mech. Behav. Biomed. Mater. 14, 7–18. Ibrahim, E., 2011. Myocardial tagging by cardiovascular magnetic resonance: evolution of techniques—pulse sequences, analysis algorithms, and applications. J. Cardiovasc. Magn. Reson. 13, 36. La Gerche, A., Ruxandra, J., Voigt, J., 2010. Right ventricular function by strain echocardiography. Curr. Opin. Cardiol. 25, 430–436. Libertiaux, V., Pascon, F., Cescotto, S., 2011. Experimental verification of brain tissue incompressibility using digital image correlation. J. Mech. Behav. Biomed. Mater. 4, 1177–1185. Lionello, G., Sirieix, C., Baleani, M., 2014. An effective procedure to create speckle pattern on biological soft tissue for digital image correlation measurements. J. Mech. Behav. Biomed. Mater. 39, 1–8. Miri, A., Barthelat, F., Mongeau, L., 2012. Effects of dehydration on the viscoelastic properties of vocal folds in large deformations. J. Voice 26, 688–697. Moerman, K., Holt, C., Evans, S., Simms, C., 2009. Digital image correlation and finite element modelling as a method to determine mechanical properties of human soft tissue in vivo. J. Biomech. 42, 1150–1153. Myers, K., Coudrillier, B., Boyce, B., Nguyen, T., 2010. The inflation response of the posterior bovine sclera. Acta Biomater. 6, 4327–4335. Pauliks, L., Valdes-Cruz, L., Perryman, R., Scholl, F., 2014. Right ventricular wallmotion changes after infant open heart surgery – a tissue Doppler study. Echocardiography 31, 209–217. Porhomayon, J., Papadakos, P., Nader, N., 2012. Failed weaning from mechanical ventilation and cardiac dysfunction. Crit. Care Res. Pract. 2012, 173527. Shimony, A., Afilalo, J., Flynn, A., 2014. Usefulness of right ventricular dysfunction to predict new-onset atrial fibrillation following coronary artery bypass grafting. Am. J. Cardiol. 113, 913–921. Sutton, M., Orteu, J.-S., JS, H., Schreier, H., 2009. Image Correlation for Shape, Motion and Deformation Measurements. Springer, New York. Teske, A., Cox, M., De Boeck, B., 2009. Echocardiographic tissue deformation imaging quantifies abnormal regional right ventricular function in arrhythmogenic right ventricular dysplasia/cardiomyopathy. J. Am. Soc. Echocardiogr. 22, 920–927. Thompson, M., Schell, H., Lienau, J., Duda, G., 2007. Digital image correlation: a technique for determining local mechanical conditions within early bone callus. Med. Eng. Phys. 29, 820–823.
Contents lists available at ScienceDirect
Journal of Biomechanics journal homepage: www.elsevier.com/locate/jbiomech www.JBiomech.com
Short communication
In-vivo deformation measurements of the human heart by 3D Digital Image Correlation Mikko Hokka a,n, Nikolas Mirow b, Horst Nagel c, Marc Irqsusi b, Sebastian Vogt b, Veli-Tapani Kuokkala a a
Tampere University of Technology, Department of Materials Science, Finland Heart Surgery, Universitätsklinikum Gießen und Marburg GmbH, Marburg, Germany c LaVision Ltd, Göttingen, Germany b
art ic l e i nf o
a b s t r a c t
Article history: Accepted 13 March 2015
Fast and accurate measurements of the kinetics and deformation of the heart during cardiac surgery can be useful for assessing the best strategies for the protection of the myocardium. While measurements based on ultrasonic technology such as the transesophageal echocardiography are rapidly developing in this direction, also other analysis methods based on optical imaging have been developed within the recent decade. The improved quality of digital cameras and increased computational power of personal computers have led to the development of deformation analysis method known as Digital Image Correlation (DIC). This paper presents preliminary results on the application of the DIC technique on analysing of the movement and deformation of the myocardial movement during a cardiopulmonary bypass surgery. The results show that the natural pattern of the heart should be sufficient for DIC, but better and more accurate results could be obtained with improved contrast conditions. DIC has a potential to be used as a sensitive tool for the surgeon to monitor the cardiac function. & 2015 Elsevier Ltd. All rights reserved.
Keywords: Digital Image Correlation Myocardium Deformation
1. Introduction Online monitoring of the movements and deformation of the ventricles and atria of the human heart may be useful when assessing different myocardial protection strategies during cardiac surgery. Recently, various imaging methods based on ultrasonic measurements, such as transesophageal echocardiography and tissue Doppler techniques (Shimony et al., 2014; Pauliks et al., 2014) have been used for analysing the movement and deformation of the heart. In addition, other experimental and numerical methods based on magnetic resonance tissue tagging and finite element simulations are being used for estimating the deformation of the heart (Ibrahim, 2011; Dorri et al., 2006). Strain and strain rate measurements of the right ventricle (RV) using echocardiography have considerably enhanced the understanding of the RV function (La Gerche et al., 2010). During the operation, however, echocardiography is not easily applicable to imaging of the (RV) or the atria. The monitoring of the RV requires advanced skill from the investigator, and all investigations are performed by an invasive transesophageal probe that occasionally induces gastrointestinal injuries and bleeding. Clinical difficulties in assessing n
Corresponding author. Tel.: þ 358 408490132 E-mail address: mikko.hokka@tut.fi (M. Hokka).
http://dx.doi.org/10.1016/j.jbiomech.2015.03.015 0021-9290/& 2015 Elsevier Ltd. All rights reserved.
the RV function using echocardiography are discussed by Teske et al. (2009). RV dysfunction can cause severe complications and even death during weaning from extracorporeal circulation (ECC) and mechanical ventilation after the cardiosurgical operation (e.g. bypass grafting, valve replacement). This is especially the case for patients who in addition to the acquired heart decease are additionally suffering from pulmonary hypertension or chronic obstructive pulmonary disease (COPD) (Porhomayon et al. 2012). Therefore, there is a need for fast and accurate analysis of the local deformation of the heart that can be used for strain and strain rate measurements of the right side of the heart. Digital Image Correlation (DIC) has been used for solving of various engineering and materials science problems within the past decades, and recently also an increasing number of medical applications have been studied using DIC. Moerman et al. (2009) studied the properties and mechanical response of artificial soft tissue using a combined modelling and DIC approach. Han et al. (2012) used DIC on ultrasonic images for diagnosis of breast cancer, whereas Libertiaux et al. (2011) studied the compressive behaviour of brain tissue using DIC. Recently, Lionello et al. (2014) developed tools for generating speckle patterns for DIC measurements of soft tissues. In 3D DIC, the cameras are observing the target from different angles, and a stereo image is reconstructed from two or more individual images using (typically) a pinhole camera model. The
2218
M. Hokka et al. / Journal of Biomechanics 48 (2015) 2217–2220
pinhole model is typically calibrated using an image of a calibration target obtained with an identical optical setup that was used for the imaging of the analysed target. The pinhole model itself is used for mapping of the two or more camera coordinate systems and the world (absolute) coordinate system. The deformation analysis is based on tracking of the movement of an image subset between consecutive images obtained as a function of time. The movement of the image subset gives the 3D displacement vectors, which can be obtained on the surface of the target at a very high spatial resolution. The tracking algorithm uses cross-correlation to find the exact location of the subset in the next image, and the deformation of the subset is taken into account using a shape function that allows the image subset to deform as a function of time. The displacements of the subsets can be obtained at a precision less than one pixel by interpolation of the grey level values between the pixels. A good overview of the DIC method, more detailed mathematical descriptions of the calibration, pinhole camera model, cross correlation, shape functions etc. as well as several examples can be found in, for example, Ref. (Sutton et al., 2009).
2. Materials and methods
positioning of the cameras were done simply by estimating where the heart of the patient would most likely be during the operation. Due to the fact that the exact location of the area of interest was not known, 50 mm lenses were used in acquiring the images, so that a large area would be covered in the images. Intraoperative image collection was carried out at the rate of 10 frames per second using LaVision E-Lite 5MPix digital cameras. The calibration of the DIC system was done by obtaining several images of the 3D calibration plate with 2.2 mm dots, 10 mm distance between the dots, and 2 mm distance between the two planes. No extra pattern was added on the surface of the myocardium and the pattern matching was carried out on the natural pattern of the surface. The DAVIS software (DaVis 8.20, LaVision Ltd) was used for obtaining the images as well as for the displacement and strain calculations. The exact mathematical formulation used for pattern matching, displacement and strain calculations and region growth algorithms are presented elsewhere (Fleet and Weiss, 2006; Bouguet, 2000). The calibration of the DIC setup had to be done prior to and/or after the surgery. Therefore, the exact distance from the calibration plate to the final/actual target was not known at the time when the calibration images were obtained. This distance was estimated by first measuring the distance from the calibration target and later on from the patient's chest to the OR floor. This estimated distance was used for the first calibration, which was then iteratively improved by transforming the calculated surface to Z¼ 0 mm position. A reasonable calibration was obtained with a standard deviation of the fit of 0.46 pixels, where one pixel corresponds to a distance of 146.8 μm.
3. Results and discussion The displacement vectors were calculated from the masked area of interest by choosing several seed points where the correlation
The experiments were carried out at the Universitätsklinikum Marburg, Germany. The image correlation system was setup in the operating theatre approximately 45 min prior to the operation. The cameras were positioned above and in front of the patient's chest, as shown in Fig. 1. However, the cameras were positioned and focused prior to bringing the patient to the operating room (OR) and therefore the focusing and
Fig. 1. Positioning of the cameras in the operating room.
Fig. 2. Stereo reconstruction errors at 0.1 s and 0.6 s corresponding to (a) average and (b) maximum errors. The images corresponding to average and maximum cases were selected by visual comparison of the images.
M. Hokka et al. / Journal of Biomechanics 48 (2015) 2217–2220
2219
Fig. 3. (a–c) are the minimum (compression) principal (Lagrange) strains on the surface of the heart overlaid on the reconstructed 3D image, and (d–f) are the maximum (tension) principal strains. The images correspond to time points (a and d) of 0.2 s, (b and e) of 0.4 s, and (c and f) of 0.6 s.
algorithm starts constructing the stereo images. The whole visible area of the heart was masked and the seed points were manually selected by trial and error method. Due to the rather poor quality of the surface pattern, a very large subset of 121 pixels with a step size of 1 pixel was used for the pattern matching. Large subsets include more image data and therefore increase the accuracy of the displacement measurement. However, the large subset size and small step size also dramatically increase the computation time. Using such a small step size will not significantly increase the spatial resolution of the method, but it was found to stabilise the area growth algorithm when analysing low quality patterns. This significantly reduces the number of required seed points. Furthermore, the high intensity focused lights used in the OR (shown in Fig. 1) resulted in glare from the wet surface of the heart. Normally a more diffused light source is preferred when recording images for DIC. The glare changes location in consecutive images and effectively prevents image correlation in areas with strong glare. In the images obtained in this work, strong glare occurred in the top parts of the images or on the lower side of the heart. Due to the low contrast of the pattern and the glare issues, most of the calculations were carried out on the surface of the right atrium, and the right ventricle was analysed only partly. Despite the poor contrast of the surface pattern and the glare problems, the stereo reconstruction errors remained relatively small and larger errors were observed only at some “hot spots”, where the pattern matching algorithm did not find a good match in the images obtained with the two cameras. Fig. 2a and b shows the 3D reconstruction overlaid with the stereo reconstruction error in pixels obtained at 0.1 s and 0.6 s from the beginning of the recording. The stereo reconstruction errors in the obtained image pairs were analysed simply by visually comparing the images where the stereo reconstruction error is overlaid with the 3D image. On average, the errors obtained in all images are very close to that shown in Fig. 2a, except for the images with maximum deformation, which are shown in Fig. 2b. The maximum errors are found repeatedly at locations with maximum deformation. At this stage the stereo reconstruction
error remains small around the atria, and the maximum errors are found at the centre of the analysed region. In this region the stereo reconstruction errors are as high as 4 pixels, which is due to the poor quality and the large deformation of the pattern. The low quality contrast patterns in analysing with DIC are a common problem for soft tissues. Gao and Desai (2009) studied porcine liver tissues using DIC with natural patterns. They carried out translation tests to evaluate the quality of the natural pattern. The sample was simply translated without any deformation and the displacements of the surface were calculated using 2D global DIC. The standard deviation of displacements for one image pair in their test varied between 0.04 and 0.44 pixels, and the cumulative standard deviation for 15 images was 0.69 pixels. The stereo reconstruction errors in the current work compare well with the values of Gao et al. but only when the deformation of the heart is small. The simple way to improve the accuracy is to improve the contrast pattern. Several research groups have made considerable efforts in improving the contrast patterns of soft tissues. Miri et al. (2012) were able to measure the deformation of porcine vocal folds to strains up to 50%. However, they used commercial tissue dyes to improve contrast of the pattern. Other groups have improved the contrast of the pattern by powders (Myers et al., 2010) or powders mixed with gels (Thompson et al., 2007). However, the improved pattern for the in-vivo measurements of the human heart remains a challenge due to the requirements for a sterile biocompatible or removable pattern. This problem needs to be addressed in future research regarding DIC analysis of cardiac function during open heart surgery. Fig. 3 shows a short time series of the DIC strain maps overlaid with the reconstructed 3D image. The images a–c show the minimum principal strain, and the images d–f show the maximum principal strain. The images were obtained at 0.1 s intervals, but only after every second image is shown. On average, the minimum principal strain varies between 0% and 10%. Clearly the region between the right atrium and the right ventricle deforms strongly, but unfortunately the pattern quality does not allow for a detailed
2220
M. Hokka et al. / Journal of Biomechanics 48 (2015) 2217–2220
analysis of this region. The maximum principal strains range on average between 5% and 10% of tensile deformation. 4. Conclusions Deformation of the heart was obtained using Digital Image Correlation of the optical images acquired intraoperatively during cardiopulmonary bypass surgery. The natural surface pattern of the myocardium was sufficient for the calculation of the 3D displacement vectors and the strains on its surface. The natural pattern was, however, not sufficient for the construction of the 3D images for the whole area of interest, and therefore not all areas could be studied. For improving the analysis and reducing the calculation time, the contrast of the natural pattern needs to be improved either by better illumination conditions or by developing a sterile biocompatible speckle pattern. Also, other methods requiring less contrast patterns, such as feature tracking algorithms might provide more use as an online tool during the surgery. Conflict of interest None. Acknowledgements Dr. Sven Curtze (Nacon – nanoAnalysis Consulting Inc.) is acknowledged for his efforts to make these measurements possible and for the encouragement to undertake these experiments. References Bouguet, J.-Y., 2000. Pyramidal Implementation of the Lucas Kanade Feature Tracker Description of the Algorithm. Intel Corporation Microprocessor Research Labs, Available online at.
Dorri, F., Niederer, P.F., Lunkenheimer, P.P., 2006. A finite element model of the human left ventricular systole. Comput. Methods Biomech. Biomed. Eng. 9, 319–341. Fleet, D., Weiss, Y., 2006. Optical flow estimation. In: Nikos, Paragious, Yunmei, Chen, Olivier D, Faugeras, Paragios, et al. (Eds.), Handbook of Mathematical Models in Computer Vision. Springer, USA. Gao, Z., Desai, J., 2009. Estimating zero-strain states of very soft tissue under gravity loading using digital image correlation. Med. Image Anal. 14, 126–137. Han, Y., Kim, D.-W., Kwon, H., 2012. Application of digital image cross-correlation and smoothing function to the diagnosis of breast cancer. J. Mech. Behav. Biomed. Mater. 14, 7–18. Ibrahim, E., 2011. Myocardial tagging by cardiovascular magnetic resonance: evolution of techniques—pulse sequences, analysis algorithms, and applications. J. Cardiovasc. Magn. Reson. 13, 36. La Gerche, A., Ruxandra, J., Voigt, J., 2010. Right ventricular function by strain echocardiography. Curr. Opin. Cardiol. 25, 430–436. Libertiaux, V., Pascon, F., Cescotto, S., 2011. Experimental verification of brain tissue incompressibility using digital image correlation. J. Mech. Behav. Biomed. Mater. 4, 1177–1185. Lionello, G., Sirieix, C., Baleani, M., 2014. An effective procedure to create speckle pattern on biological soft tissue for digital image correlation measurements. J. Mech. Behav. Biomed. Mater. 39, 1–8. Miri, A., Barthelat, F., Mongeau, L., 2012. Effects of dehydration on the viscoelastic properties of vocal folds in large deformations. J. Voice 26, 688–697. Moerman, K., Holt, C., Evans, S., Simms, C., 2009. Digital image correlation and finite element modelling as a method to determine mechanical properties of human soft tissue in vivo. J. Biomech. 42, 1150–1153. Myers, K., Coudrillier, B., Boyce, B., Nguyen, T., 2010. The inflation response of the posterior bovine sclera. Acta Biomater. 6, 4327–4335. Pauliks, L., Valdes-Cruz, L., Perryman, R., Scholl, F., 2014. Right ventricular wallmotion changes after infant open heart surgery – a tissue Doppler study. Echocardiography 31, 209–217. Porhomayon, J., Papadakos, P., Nader, N., 2012. Failed weaning from mechanical ventilation and cardiac dysfunction. Crit. Care Res. Pract. 2012, 173527. Shimony, A., Afilalo, J., Flynn, A., 2014. Usefulness of right ventricular dysfunction to predict new-onset atrial fibrillation following coronary artery bypass grafting. Am. J. Cardiol. 113, 913–921. Sutton, M., Orteu, J.-S., JS, H., Schreier, H., 2009. Image Correlation for Shape, Motion and Deformation Measurements. Springer, New York. Teske, A., Cox, M., De Boeck, B., 2009. Echocardiographic tissue deformation imaging quantifies abnormal regional right ventricular function in arrhythmogenic right ventricular dysplasia/cardiomyopathy. J. Am. Soc. Echocardiogr. 22, 920–927. Thompson, M., Schell, H., Lienau, J., Duda, G., 2007. Digital image correlation: a technique for determining local mechanical conditions within early bone callus. Med. Eng. Phys. 29, 820–823.