Australas Phys Eng Sci Med (2015) 38:223–228 DOI 10.1007/s13246-015-0339-4
SCIENTIFIC PAPER
The design of 3D scaffold for tissue engineering using automated scaffold design algorithm Shahenda Mahmoud1 • Ayman Eldeib1 • Sherif Samy1
Received: 27 August 2014 / Accepted: 10 March 2015 / Published online: 17 March 2015 Ó Australasian College of Physical Scientists and Engineers in Medicine 2015
Abstract Several progresses have been introduced in the field of bone regenerative medicine. A new term tissue engineering (TE) was created. In TE, a highly porous artificial extracellular matrix or scaffold is required to accommodate cells and guide their growth in three dimensions. The design of scaffolds with desirable internal and external structure represents a challenge for TE. In this paper, we introduce a new method known as automated scaffold design (ASD) for designing a 3D scaffold with a minimum mismatches for its geometrical parameters. The method makes use of k-means clustering algorithm to separate the different tissues and hence decodes the defected bone portions. The segmented portions of different slices are registered to construct the 3D volume for the data. It also uses an isosurface rendering technique for 3D visualization of the scaffold and bones. It provides the ability to visualize the transplanted as well as the normal bone portions. The proposed system proves good performance in both the segmentation results and visualizations aspects. Keywords Tissue engineering Scaffolds Bone defect Automated design Medical imaging
Introduction Bone is a dynamic and highly vascularized tissue that continues to remodel throughout the life time. It is a remarkable organ playing key roles in critical functions in & Shahenda Mahmoud
[email protected] 1
Department of Systems and Biomedical Engineering, Cairo University, Giza, Egypt
human physiology, including protection, movement and support of other critical organs, mineral storage and blood production, and others [1]. Bone loss is a major health care problem worldwide and representing 10 % of annual health care expenditures. It could happen due to surgery, trauma, primary tumor resection, disease or normal aging, which very often entails the use of substitutes to help repair or replace the damaged or diseased tissue. The currently available bone treatment of bone defects, have been largely centered on replacing the lost bone with tissues using autografts (patient’s own), allograft (human other than patient), or xenograft (from different species), and synthetic materials. Were autograft has the best clinical outcome as it integrates reliably with host bone and lacks the Immune and disease related complications of allograft and xenograft. Nevertheless, its use is severely hampered by its short supply, additional surgery and the considerable donor site morbidity associated with the harvest. In addition, the existence of non-living synthetic materials in the body calls for recurring surgeries since the material does not grow or adapt to the changing environments that the patient undergoes [2–4]. That way there is a growing need and demand for living substitutes that can help grow the patient’s own functional tissue. Approximately three decades ago a new alternative approach to tissue and organ reconstruction emerges; that is tissue engineering (TE). It is based on the understanding of tissue formation and regeneration, and aims to induce new functional tissues, rather than just to implant new spare parts [5–8]. One decisive factor to the success of TE strategies is the appropriate design of the scaffold; that effectively serves as man-made extracellular matrices. Ideally, a scaffold should have several characteristics: (i) provide the optimal microenvironment for cells to attach, and guides the overall shape; (ii) three-dimensional
123
224
(3D) and highly porous with an interconnected pore network for cell growth, transport of nutrients and metabolic waste; (iii) biocompatible, biodegradable; and (iv) possess proper mechanical and physical properties to match those of the tissues at the site of implantation [9–11]. In addition to these essential properties the external size and shape of the construct must also be considered, especially if the construct is customized for an individual patient to prevent any internal or external hazards [12]. A recent design and manufacture methodology rapid prototyping (RP) [13] includes a number of layer by layer manufacturing processes that enable complex 3D anatomic scaffold architectures to be built using computer-aided design (CAD) techniques and data from patient scans [14]. Design of scaffolds in current enabling CAD software is by no means an easy task and extremely time consuming [2, 15, 16]. The objective of this paper is to introduce a new method for designing, known as automated scaffold design (ASD). It aims to design 3D scaffolds, with a minimum mismatch for its geometrical parameters including a novel internal architecture design. It also contains model structure for simpler scaffold shape for experimental purpose. The paper briefly explains the methods which are used in ASD for segmentation, registration and 3D rendering visualization. The proposed system proves good performance in both the segmentation results and visualizations aspects.
Materials and methods Data acquisition The system methodology was performed with MATLAB (2009). The algorithm is tested with two-dimensional (2D) cross-section images (axial cuts) from CT acquired by a (GE medical systems) CT scanner. After acquisition of CT image data is recorded in digital imaging and communications in medicine (DICOM) standard format, the most common in medical image management. In DICOM files, detailed information on scanning parameters is recorded too. Here the scanning interval between images for different cases between range of (2–3.5 mm) and image resolution of 512 9 512. The image should contain both legs of the patient and only one of them is defected. Segmentation of slice images Segmentation is the process of classifying pixels in an image or volume. It is one of the most difficult tasks in the visualization processes [17, 18]. For reconstruction of medical 3D surface and volume, interest bone tissue boundaries should be distinguished from others on all the image slices. After CT scanning, on the slice images,
123
Australas Phys Eng Sci Med (2015) 38:223–228
scanner table’s images are also appearing. So, during automatic segmentation, table pixels might frequently be classified as bone tissue pixels and this situation is undesired for accurate segmentation and 3D reconstruction. In order to overcome this problem, table images should be extracted and eliminated from the images. In other words, only the volume of interest (VOI) should be residing on the images. To select VOI from each slice image, ASD uses filtering techniques to eliminate the undesired data. Automated scaffold design performs segmentation using k-means algorithm. The k-means clustering is an iterative technique that is used to partition an image into K clusters [19, 20]. ASD algorithm partitions the image into six clusters reflecting the different parts. The six clusters are (edge, background, fatty tissues, muscles, bone, and defect). This algorithm aims at minimizing an objective function, in this case a squared error equation: k X n 2 X ðjÞ J¼ ð1Þ xi cj j¼1 i¼1
2 ðjÞ where xi cj is a chosen distance measure between a data point xi (j) and the cluster center cj, is an indicator of the distance of the n data points from their respective cluster centers. Then, ASD applies morphological gap filling and closing to the output [17, 21]. At the end of this phase we should have a set of images containing the patient bone only. Registration of 2D slice images Registration is the procedure of mapping points from one image to corresponding points in another image in order to monitor subtle changes between the two images [22]. The target of registration is to determine the difference between the healthy bone and the defected one in order to design a 3D scaffold model for the output. Before entering the registration phase ASD splits each image into two images, one contain the healthy bone and the other contain the defected one. ASD divides registration into three stages. First stage (transformation and translation) purpose is to align the healthy and defected bone per slice on each other to determine the differences between them, which represent the missing bone (Fig. 1). This process is done using affine transformation which is sufficient to match two images of a scene taken from the same viewing angle but from a different position [23]. The alignment set parameters are simply composed of (rotation, translation and scaling), which can be expressed using the composite matrix in the following equation: 2 03 2 32 3 x s cos h s sin h tx x 4 y0 5 ¼ 4 s sin h s cos h ty 54 y 5 ð2Þ 1 0 0 1 1
Australas Phys Eng Sci Med (2015) 38:223–228
225
Fig. 2 For the same slice. a The health, bone image. b The defected bone image. c The difference between a and b
Fig. 1 An alignment performed by ASD where the red area represents the healthy bone, the green area represents the defect bone and the yellow area represent the common area between them
where h is the angle of rotation, s scaling factor and (tx, ty) are translation vector. In the second stage (mean square error) and at each transformation parameter set the mean square error (MSE) [24], is calculated using the following equation: M X N 1 X 2 ðf ðx; yÞ f 0 ðx; yÞÞ MN y¼1 x¼1
ð3Þ
where f (x, y) is the one’s pixels of the healthy bone image, f0 (x, y) is the one’s pixels of the defected one and M, N are the dimensions of the images. By taking the positive difference, the parameter set, which gives the minimum error or minimum deviation per slice is selected as the candidate transformation parameters and used to transform one image to the other. If it happen and two parameters set with the same least MSE per slice, then the parameter set which is compatible with the previous set is selected as the candidate. Third stage (determine the difference) is for determining the difference between two images per sliced after alignment using the following equation: 1 f ðx; yÞ f 0 ðx; yÞ [ 0 f ðx; yÞ ¼ ð4Þ 0 otherwise If the existing difference is greater than zero, it will be taken into the account, otherwise it won’t be considered. By ending this sub-phase the difference output which represents the defected area is identified (Fig. 2c). 3D Rendering of registered data The 3D rendering is the process that leads to the visualization of 3D object [25]. ASD 3D visualization is obtained using the Matlab function isosurface, which implements Lorensen ‘‘Marching Cubes’’ algorithm [26]. This algorithm produces a triangle mesh by computing isosurfaces from discrete data. We reach a surface representation, by connecting the patches from all cubes on the isosurface boundary. The algorithm
Fig. 3 Two different patients’ cases. a First case, defected bone. b The scaffold shape representing only the defected area in the first case. c Second case, defected bone. d The scaffold shape representing the healthy bone equivalent to the defected area in second case due to huge deformation of the bone
operates through the scalar field, by taking eight neighbor locations at a time (thus forming an imaginary cube), then determining and deciding the polygon(s) needed to represent the part of the isosurface that passes through this cube [26, 27]. The technique of 3D rendering is used to create a 3D model to represent (i) both legs together, (ii) the defected area, (iii) either the healthy bone or the defected one separately, (iv) the scaffold outer and inner shape, (v) the transplantation of the scaffold in the defected area.
Result Sample scaffold design with (ASD) To construct the desired scaffold model, first the external architecture shape for the scaffold need to be determined.
123
226
Australas Phys Eng Sci Med (2015) 38:223–228
Fig. 4 Different designed shapes for the scaffold inner unit cell
Fig. 6 For two different cases. a, d 3D Visualization of defect bone. b, e 3D visualization of the erosion. c, f 3D visualization of erosion transplant
Fig. 5 For the same case. a The scaffold inner unit cell intersects with scaffold outer shape. b The installation of the erosion with the defected bone. c Using ASD sub volume mode
According to the patient case, the external output shape desired is either to be represented by the defected parts only (Fig. 3b) or by the healthy bone equivalent to the whole area (Fig. 3d). Second, ASD designs the scaffold unit cell (internal architecture), and pores size for the desired bone tissue by selecting the shape and determining the size of the selected unit cell. ASD gives the ability to design specific pore geometry for pore size and shape (Fig. 4). Then, arranging feature patterns in a specific 3D architecture to form a preferable pore distribution and interconnectivity (for cell growing and proliferation) to meet the scaffold strength and stability requirements. After that, the scaffold inner unit cell intersects with scaffold outer shape to have the finial scaffold model (Fig. 5a). ASD has a few methods to see and imagine the final output shape of the scaffold. The first method is the
123
Fig. 7 Different shapes and view for scaffold model blocks
transplantation of the scaffold representing the erosion into the defected bone. This happens by adding or intersecting between the scaffold and the defected bone (Figs. 5b, 6). This method helps us to ensure that the scaffold fits the area, it needs to transplant in specially if the scaffold is customized for a certain patient. The second method is by using the slicing mode. This mode extracts subset of a
Australas Phys Eng Sci Med (2015) 38:223–228
227
Fig. 8 The quality measurement of the system segmentation, where SE represents sensitivity, SP represent specificity and ACC represent accuracy. a The output measurement for 25 images of healthy bone. b The output measurement of 25 images of defected bone
Table 1 Comparison of 3D reconstruction systems
Software
Mimics
AutoCAD
ASD
Price
Very expensive
Cheap
Very cheap
Quality of reconstruction
High quality
High quality
Fairly high quality
Type of segmentation
Automatic
Manual
Automatic
Speed of reconstruction
High
Slow
Fairly high
volume data set, which gives the ability to see the grid construction inside the scaffold (Fig. 5c). Another ASD mode is to facilities the ability to intersect the selected unit cell with structure model (simple block) scaffold for experimental use (Fig. 7). At the end a desirable 3D scaffold model is designed.
Discussion and conclusion One of the main pillars of ASD system is segmentation, which is a key step in determining the external geometrical shape of the scaffold. And to ensure that the segmentation has the required performance, we need to measure the
123
228
sensitivity, specificity and accuracy [28] of the system by collecting ground truth set of segmentation data (reference data) and compare it with the system segmentation output (actual data). For preparing the reference data, we used a semi-automatic medical image segmentation system ‘‘Livewire’’ also known as intelligent scissors. By applying this test on 50 slices split between 25 images of healthy bones and 25 images of defects, bone. We were able to get the following results quality measures (Fig. 8). The analysis of the segmentation measures and the high values of the accuracy measures approve that the algorithm handles a very good performance regard the segmentation of the bone tissues. Another pillar in the ASD system is the 3D visualization or the regeneration of 3D model for the entered data. To evaluate the performance of ASD system regards this point the paper present a short comparison between ASD system and another 3D reconstructing surface systems, e.g. Mimics and AutoCAD as illustrated in (Table 1). One the important advantage of ASD system is the automated segmentation and the price, which is much less than the other two methods. And one of the limitations of the system is the quality of reconstruction level not as high as the other two systems but still fairly high. The ASD system proposed in this paper facilitates the creation of scaffold designs from the assigned unit cell and the desired scaffold output shape. The advantages of this design process include: (i) Good segmentation performance, which allow having the dispersed outer shape of the scaffold. (ii) The 3D visualization of the defect as well as the normal bone portions. (iii) The design of outer scaffold shape and the design of its interior are separated and hence do not present a memory overhead. (iv) The Facilitation to see the output scaffold transplant. (v) Data sub-volume, which enables to see the grid construction inside the scaffold. (vi) Use of model block for scaffold outer shape dispensing the necessity of medical images and complicated shapes for experimental studies. Acknowledgments This work is supported by Department of Systems and Biomedical Engineering, Faculty of Engineering, Cairo University, Egypt.
References 1. Porter JR, Ruckh TT, Popat KC (2009) Bone tissue engineering: a review in bone biomimetic and drug delivery strategies. Biotechnol Prog (AIChE) 25:1539–1560 2. B. Starly (2006) Biomimetic design and fabrication of tissue engineered scaffolds using computer aided tissue engineering. A thesis submitted to Drexel University, USA 3. Stevens MM (2008) Biomaterials for bone tissue engineering. Mater Today 11:18–25. doi:10.1016/S1369-7021(08)70086-5
123
Australas Phys Eng Sci Med (2015) 38:223–228 4. Silber JS et al (2003) Donor site morbidity after anterior iliac crest bone harvest for single-level anterior cervical discectomy and fusion. Spine 28:134–139 5. Bose S, Roy M, Bandyopadhyay A (2012) Recent advances in bone tissue engineering scaffolds. Trends Inbiotechnol 30(10):546–554 6. Kneser U, Schaefer DJ, Munder B, Klemt C, Andree C, Stark GB (2002) Tissue engineering of bone. Minim Invasive Ther Allied Technol 11(3):107–116 7. Langer R, Vacanti JP (1993) Tissue engineering. Science 260(5110):920–926 8. Ikada Y (2006) Challenges in tissue engineering. J R Soc Interface 3:589–601 9. Bonadio J, Smiley E, Patil P, Goldstein S (1999) Localized, direct plasmid gene delivery in vivo: prolonged therapy results in reproducible tissue regeneration. Nat Med 5(7):753–759 10. Hutmacher DW (2000) Scaffolds in tissue engineering bone and cartilage. Biomaterials 21(24):2529–2543 11. Hutmacher DW (2001) Scaffold design and fabrication technologies for engineering tissues state of the art and future perspectives. J Biomater Sci Polym Ed 12(1):107–124 12. De Boer J, Van Blitterswijk C, Thomsen P, Hubbell J, de Ranieri Cancedda JD, Bruijn AL, Sohier J, Williams DF (2008) Tissue engineering, 1st edn. Elsevier Academic Press, London 13. Salgado AJ, Coutinho OP, Reis RL (2004) Bone tissue engineering: state of the art and future trends. Macromol Biosci 4:743–765 14. Hollister SJ (2005) Porous scaffold design for tissue engineering. Nat Mater 4(7):518–524 15. Starlya B, Laub W, Bradburyb T, Sun W (2005) Internal architecture design and freeform fabrication. Comput Aided Des 38:115–124 16. Ramin E, Harris RA (2009) Automated design of tissue engineering scaffolds by advanced cad. Proc Inst Mech Eng 223(3):289–301 17. Gonzalez RC, Woods RE (2002) Digital image processing. Prentice-Hall, Englewood Cliffs 18. Singh KK, Singh A (2010) A study of image segmentation algorithms for different types of images. Int J Comput Sci Issues (IJCSI) 7(5):414–417 19. Jipkate MBR, Gohokar VV (2012) A comparative analysis of fuzzy c-means clustering and k means clustering algorithms. Int J Comput Eng (IJCER) 2(3):737–739 20. Ghosh S, Dubey SK (2013) Comparative analysis of k-means and fuzzy c-means algorithms. Int J Adv Comput Sci Appl (IJACSA) 4:35–39 21. Umbaugh SE (1998) computer vision and image processing. Prentice Hall, Upper Saddle River 22. Antoine Maintz JB, Viergever MA (1998) A survey of medical image registration. Med Image Anal 2:1–36 23. Deshmukh MP, Bhosle U (2011) A survey of image registration. Inter J Image Process (IJIP) 5:245–269 24. Kumar R, Rattan M (2012) Analysis of various quality metrics for medical image processing. Inter J Adv Res Comput Sci Softw Eng(IJARCSSE) 2(11):137–144 25. Torres PMB, Gonc¸alves PJS, Martins JMM (2012) 3D Reconstruction and visualization of femur bone structures. Rom Rev Precis Mech Opt Mechatron 41:51–56 26. Lorensen WE, Cline HE (1987) Marching cubes: a high resolution 3d surface construction algorithm. ACM SIGGRAPH Comput Graph 21(4):163–169 27. Kalyankar PP, Apte SS (2013) 3D Volume rendering algorithm. Int J Eng Adv Technol (IJEAT) 2(5):268–270 28. Wikipedia website, en.wikipedia.org/wiki/Sensitivityand specificity
