Scholarly article on topic 'Development of patient specific implants for Minimum Invasive Spine Surgeries (MISS) from non-invasive imaging techniques by reverse engineering and additive manufacturing techniques'

Development of patient specific implants for Minimum Invasive Spine Surgeries (MISS) from non-invasive imaging techniques by reverse engineering and additive manufacturing techniques Academic research paper on "Medical engineering"

CC BY-NC-ND
0
0
Share paper
Academic journal
Procedia Engineering
OECD Field of science
Keywords
{"Point Cloud Data" / "CT scans" / "Image Processing" / threshold / NURBs / "free form surface"}

Abstract of research paper on Medical engineering, author of scientific article — V.N. Chougule, A.V. Mulay, B.B. Ahuja

Abstract Reverse Engineering and Rapid Prototyping are extensively used technologies by both research and industrial community for rapid developments in various industrial as well as Bio-medical applications. Recent advances in computer technology and Bio-medicines enabled Computer Aided Design (CAD) to find many novel applications in Bio-medical engineering and integration of CAD and Bio-medical technology is usually referred as Bio-CAD. The major objective of the current research work is to generate an efficient algorithm for generation of free form surface from non-invasive CT scan images. Minimally Invasive Spine Surgery (MISS) have enabled spinal surgeons to select patients and treat several spinal disorders like degenerative disc disease, herniated disc, fractures, tumors, infections, instability, deformity, etc. with less disruption of muscles, which enables patient towards faster recovery to normal functions, reduces operative blood loss. In this paper, it is proposed to extract point cloud data from stalk of non-invasive CT scan images by using Image Processing techniques and Reverse Engineering approach. This point cloud data is to be processed for noise reduction, point cloud data segmentation and CAD model generation. This image-based CAD modeling approach begins with the acquisition of CT scan in DICOM 3.0 format. The point cloud estimation is based on threshold techniques and edge detection method. This point cloud data is used for construction of 3D CAD model by fitting free form NURB surface between theses points and then fitting surface between these curve networks by swept blend technique. An efficient and robust algorithm has been developed for generation of curves from unorganized point cloud data.

Academic research paper on topic "Development of patient specific implants for Minimum Invasive Spine Surgeries (MISS) from non-invasive imaging techniques by reverse engineering and additive manufacturing techniques"

CrossMark

Available online at www.sciencedirect.com

ScienceDirect

Procedía Engineering 97 (2014) 212 - 219

Procedía Engineering

www.elsevier.com/locate/procedia

12th GLOBAL CONGRESS ON MANUFACTURING AND MANAGEMENT, GCMM 2014

Development of patient specific implants for Minimum Invasive Spine Surgeries (MISS) from non-invasive imaging techniques by reverse engineering and additive manufacturing techniques

V. N. Chougulea*, A. V. Mulayb, B. B. Ahujab

aDepartment of Mechanical Engineering, M.E.S. College of Engineering, Pune, MH, INDIA bDepartment of Production Engineering, College of Engineering, Pune, MH, INDIA

Abstract

Reverse Engineering and Rapid Prototyping are extensively used technologies by both research and industrial community for rapid developments in various industrial as well as Bio-medical applications. Recent advances in computer technology and Bio-medicines enabled Computer Aided Design (CAD) to find many novel applications in Bio-medical engineering and integration of CAD and Bio-medical technology is usually referred as Bio-CAD. The major objective of the current research work is to generate an efficient algorithm for generation of free form surface from non-invasive CT scan images. Minimally Invasive Spine Surgery (MISS) have enabled spinal surgeons to select patients and treat several spinal disorders like degenerative disc disease, herniated disc, fractures, tumors, infections, instability, deformity, etc. with less disruption of muscles, which enables patient towards faster recovery to normal functions, reduces operative blood loss. In this paper, it is proposed to extract point cloud data from stalk of non-invasive CT scan images by using Image Processing techniques and Reverse Engineering approach. This point cloud data is to be processed for noise reduction, point cloud data segmentation and CAD model generation. This image-based CAD modeling approach begins with the acquisition of CT scan in DICOM 3.0 format. The point cloud estimation is based on threshold techniques and edge detection method. This point cloud data is used for construction of 3D CAD model by fitting free form NURB surface between theses points and then fitting surface between these curve networks by swept blend technique. An efficient and robust algorithm has been developed for generation of curves from unorganized point cloud data.

© 2014TheAuthors.PublishedbyElsevierLtd.Thisis an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/3.0/).

Selection and peer-review under responsibility of the Organizing Committee of GCMM 2014 Keywords: Point Cloud Data; CT scans; Image Processing; threshold; NURBs; free form surface

* Corresponding author. Tel. : 9420195928 E-mail address: vnchougule@mescoepune.org

1877-7058 © 2014 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/3.0/).

Selection and peer-review under responsibility of the Organizing Committee of GCMM 2014 doi: 10.1016/j.proeng.2014.12.244

1. Introduction

Due to rapid advancement in information technology and reverse engineering techniques, reconstruction of three dimensional (3D) Bio-CAD models from computer tomography (CT) images has recently become the issue of much attention. With the help of medical imaging and free-form technologies like Reverse Engineering (RE) and Rapid Prototyping (RP) has the capacity to build anatomic models which have diagnostic, therapeutic and rehabilitatory medical applications. These Bio-CAD models are usually very handy for medical practitioners in numerous Bio-medical applications viz. computer-aided surgery, structural modelling of tissue, design of orthopaedic device, implants, tissue scaffolds and freeform fabrication or bio-manufacturing [1-4]. These models also can be much useful in some non-medical applications like passenger safety design and crash analysis [2, 4-7].

Largely the volume based or contour based approaches are used for the reconstruction of 3D Bio-CAD models from medical images. For interpretation of CT scan data, voxel (i.e. volume pixel) representation is most commonly used in mainstream commercially available Bio-medical software. Voxel model is represented by cubical or prismatic elements with a set of height, width and depth. These models can be directly transferred to stl (stereo-lithography) format and printed on a rapid prototyping system. It generates high resolution surfaces for visualization of human anatomy, but suffers from some limitations like huge memory requirements, loss of geometric information and inconvenient for model editing. Also, triangular patched model from the marching cube is difficult for manufacturing and engineering applications without pre-processing [8, 10]. These limitations of voxel based models can be superseded by converting CT image models into CAD based solid models by using contour based approach. CAD based models are represented by the boundaries that enclose it in 'boundary representation' techniques. Polynomial functions viz. B-Spline curves, Non-Uniform Rational B-Spline (NURBS) functions are used to define bounding surfaces. It provides additional advantage of achieving higher degree continuities viz. Tangential (C1) and Curvature (C2) over Positional (C0) continuity in stl data. This method provides advantage over volume based approach due to editable and reduced file sizes and ensuring water tight surface models.

The Reverse Engineering (RE) techniques can be employed through steps viz. capture, pre-processing for noise/out-liars removal, segmentation and model generation from point cloud data [8,9]. Human anatomical data can be attained from 2 dimensional non-invasive techniques like Computer Tomography (CT) / Magnetic Resonance Imaging (MRI) in DICOM format which are very popular techniques in medical diagnostics and surgical planning. By using contemporary computer graphics and scanning technologies, generation of 3D models is possible for visualization or modelling [10,11]. The reconstructed 3D models can provide valuable medical information and powerful diagnostic tool for surgeons to understand the complex internal anatomy of the patient. These can also be used to fabricate prostheses and to perform various simulation and analytical tasks [1,8,10,11].

This paper deals with generation of Point Cloud Data (PCD) by processing Digital Imaging and Communications in Medicines (DICOM) using Image Processing techniques from non-invasive medical images viz. CT / MRI scans. Image Processing techniques are proposed to extract point cloud data from stalk of CT scan images. Point Cloud data is data is unevenly spaced, contains lot of noise and the shape, it represents is full of intricate details, which are not present in conventional CAD data. These points are further pre-processed for sorting, smoothening and B spline curve fitting. In pre-processing of these point cloud data is sorted to differentiate internal as well as external boundaries, smoothening. Further it is proposed to fit B-spline curves through these points. This method enables the construction of the CAD model by reducing file size and ensuring the construction of water tight surfaces. The volumetric data i.e. voxel and triangular facet triangle based models are primarily used for bio-modelling and visualization, which requires huge memory space. On the other side, recent advances in CAD technology facilitate design, prototyping and manufacturing of any object having freeform surfaces. These CAD-based solid modelling is based on boundary representation (B-rep) techniques. This data is converted into 3D CAD model and in turn printed using Rapid Prototyping.

Nomenclature

DICOM Digital Imaging and Communications in Medicines

PCD Point Cloud Data

NURBS Non-Uniform Rational B-Spline

C0, C1, C2 Positional, Tangential and curvature Continuity

CT Computer Tomography

MRI Magnetic Resonance Imaging

stl Stereo-lithography

HU Hounsfield Units

STP Standard temperature and pressure

2. Methodology CT images

Fig. 1. Methodology for conversion of CT scan images to3D CAD model

Based on extensive literature survey, the methodology adopted is as shown in Fig. 1. DICOM images stores the numerical information of each pixel as CT number or Hounsfield number. The Hounsfield unit (HU) scale is a linear transformation of the original linear attenuation coefficient measurement into one in which the radio density of distilled water at standard pressure and temperature (STP) is defined as 0 HU, while the radio density of air at STP is defined as -1000 HU. For current research work, images have been acquired from GE ProSpeed CT scanner in

DICOM 3.0 format. This scanner uses a Hounsfield Unit scale ranging from -1500HU to +4000 HU. HU value varies from +700 HU for cancellous bones to +3000 HU for dense bone [20]. Target is to achieve a point cloud data from these set of DICOM images by using automatic thresholding technique. A novel programming module is designed to get these point cloud data from input DICOM images. Very large amount of point cloud data is generated from these images. These data requires pre-processing for noise reduction due to poor quality images, implants, noise filtering, smoothening of data, etc. For faster processing of PCD, segmentation of data is required.

3. Conversion of CT scan images to CAD model

Most of the current research work involves in collecting the feature points from images using spatial domain methods. Such information includes edges, interest or corner points, ridges etc. To isolate region or object of interest from background, segmentation techniques are used. In current work image segmentation is used to separate the bone from surrounding soft muscles in set of CT images. For image based segmentation, threshold based techniques are widely accepted by researchers [3,4,9,12,16-20]. These techniques can be broadly classified as global and local thresholding. In global thresholding, threshold values can be manually or automatically determined and classified based on its intensity values. Thresholding based techniques can be used to separate out bones from its soft tissues due to high intensity levels or HU values of bones in CT images. Global thresholding is very sensitive to partial volume effects, high grey level intensity of surrounding pixels due to implants, beam hardening, etc.

Local thresholding methods are based on determination of a different threshold value of each pixel based on its local statistics. Most popularly used techniques are the mean of the maximum and minimum pixel intensities [3,4], the mean plus standard deviation or median of pixel intensities [9,12], statistics based on local intensity gradient magnitude [9]. These techniques are also susceptible for partial volume effect and intensity inhomogeneity. Wang et al used the simplified Kang method to validate bone segmentation with empirically chosen threshold value with optional manual interaction to refine the resulting periosteal surface[13]. Maneka R. et al proposed use of Discrete Curvelet transformation and Features from Accelerated Segmentation Test (FAST) [3, 4]. These approaches rely on initial thresholding to extract only the bony areas from each slice required for feature point extraction. Further Curvelet Transformation is applied to detect the edges from threshold slices.

Fig.2 Methodology for generation of Point Cloud Data

Fig. 2 represents methodology adopted for generation of Point Cloud Data from CT scan images in DICOM 3.0 format by using automatic thresholding technique. A novel programming module is developed in MATLAB to get these point cloud data from input DICOM images. The principal source of data is CT images retrieved from GE

ProSpeed CT scanner in DICOM 3.0 format. These grayscale images had been acquired from GE ProSpeed CT scanner in DICOM 3.0 format of 512 X 512 pixel size with data collection diameter 250 mm with slice thickness of 5 mm.

Fig.3. Stepwise results- Image Processing

Fig. 3 shows stepwise results obtained from developed image processing module. From single CT scan slice, 2121 points are extracted. These points are stored as (x, y, z) coordinates in neutral file format 'ASCII' which can be further used in any CAD environment. This case study represents the first step of wider experimental protocol of conversion of CT images into 3D CAD model, which will provide higher order continuity (C2) as well as provide additional advantages of CAD models like editing, etc. A new threshold based algorithm is designed for extraction of Point Cloud Data from CT scan. Since, bone is having heterogeneous structure, with variable density; estimation of threshold value is most important step in this algorithm. The proposed use of these point cloud data is to construct B-Spline curves and surfaces which will enable us to achieve Tangential (C1) and Curvature (C2) continuity. For rapid prototyping, this point cloud data can be transferred to *.stl file with only positional continuity (C0). The major challenge in this work is huge number of points, to distinguish between inner and outer bone boundary profile points and detection of nearest points on one surface. Fig. 4(a) shows results obtained from CT scan of lumbar spine of 5 year female patient containing 330 slices with 1 mm slice thickness resulting into 11,59,752 points. Fig. 4(b) indicates point cloud data of 26,564 points from human skull CT scan of 59 year old male patient with 11 slices of slice thickness 5 mm.

Fig. 4. Point Cloud Data (a) Spine; (b) Skull.

For fitting 3D point cloud data, various surface reconstruction algorithms for surface fitting like loft surface, sweep surface, B spline surface and quadratic surface fitting have been proposed by researchers [11,12,20]. To achieve Curvature Continuity (C2), minimum three degree curve must be fitted through point cloud. Generally, for having three degree curve, Cubic splines, B-spline and Non-Uniform Rational B-Spline (NURBS) can be used. Among them, B-spline surfaces, especially NURBs surfaces, are gaining more popularity due to their ability to accurately approximate most types of surface entities encountered in design and manufacturing application. Dong-Jin Yoo proposed an efficient implicit surface interpolation scheme to reconstruct a B-spline surface from a point cloud data or a sequence of CT image data [12]. To maintain surface accuracy in the CAD model, Closed NURBS curves can be generated along contours of each layer. Olya Grove et al proposed use of medical image data to reconstruct a surface from serial parallel contours extracted from images [20]. B spline curves are fitted through the points along contours of the area of interest. These curves can be lofted to generate 3D surface model. These control point based methods viz. B-splines and NURBs provides excellent coverage but sometimes uncontrollable due to high degree of freedom resulting from large number of control points. These curves provide higher order continuities as compared to 'stl' and voxel based methods.

Fig.5. (a) Point Cloud Data; (b) B-spline curve through PCD

For curve fitting through sizable point cloud data, interpolation and approximation are two most commonly used methods. In interpolation, curve passes through all measured data points. This method generally fails due to too many wiggles in interpolating curves that were not removable even with a high tolerance. Unlike interpolation, in approximation, curve approximates through data points and smoothness depends upon degree of curve. It fails due to failure of estimating degree of freedom, also it loses intricate shapes. To improve the modelling process, preprocessing of Point Cloud Data for removal of outliers and smoothening is required. Fig. 5(a) shows a sample Point Cloud Data and Fig.5(b) shows a closed B-spline curve fitted through sample point cloud data.

Fig.6. (a) Original CT Scan; (b) Enhanced CT image; (c) Point Cloud Data; (d) NURB's Model; (e) RP Model; (f) Removed piece of bone from vertebra after surgery

Refer Fig. 6 for results obtained from CT scan with 1 mm slice thickness of 5 year female patient suffering from diastematomyelia. Fig. 6(a), (b) and (c) represents sample CT scan in DICOM format, enhanced image after conversion in jpeg format and noise removal by using image processing techniques and point cloud data generated from CT scan images in region of interest in Imageware software respectively. Further these point cloud data is processed for removal of outliers and smoothening and NURBs model is constructed from point cloud data as shown in Fig. 6(d). Physical model of vertebra is generated by using Stratesys Dimension 768 fused deposition modelling (FDM) machine as shown in Fig. 6(e). Fig. 6(f) shows piece of bone removed after surgery, which confirms with estimated size of bone or thorn like bony structure developed in vertebra.

4. Conclusion:

In this paper discussed steps of wider experimental protocol of conversion of CT images into 3D CAD model, which will provide higher order continuity (C2) as well as provide additional advantages of CAD models like editing, etc. A new threshold based algorithm is designed for extraction of Point Cloud Data from CT scan. By using Image Processing techniques for noise reduction and edge detection, edges were identified, which is further converted into (x, y, z) coordinates and data stored in ASCII file, which can be further used in any CAD environment. The major challenge in this work is huge number of points, to distinguish between inner and outer bone boundary profile points, detection of nearest points on one surface. These point cloud data after pre-processing for noise reduction, smoothening and sorting can be further utilized for curve network generation through the construction of B-Spline curves which will enable us to achieve Tangential (C1) and Curvature (C2) continuity. Point cloud data is segmented and fitted with NURBs surface patches. For rapid prototyping, this point cloud data is transferred to *.stl file with only positional continuity (C0). Results are confirmed after surgery on 5 year female.

5. Acknowledgement:

This research is supported by Rajiv Gandhi Science & Technology Commission Scheme - Biomedical Engineering and Technology (incubation) Centre (BETiC) Project. The authors would like to thank Dr. Sunil M. Nadkarni, Sergeon, B. K. L. Walawalkar Hospital, Ratnagiri for providing CT scans and valuable information regarding human anatomy and surgical processes.

6. References:

[1] Kineri, Y., Wang, M., Lin, H., Maekawa, T., "B-Spline surface fitting by iterative geometric interpolation/approximation algorithms", Computer-Aided Design44, pp. 697-708 (2012).

[2] Jin, G.Q., Li, W. D., Tsai, C. F., Wang, L., "Adaptive Tool-Path Generation of Rapid Prototyping for Complex Product Models", Journal of Manufacturing Systems 30, pp. 154-164(2011).

[3] Menaka, R., Chellamuthu, C., Karthik, R., "Efficient 3D Point Cloud Generation from Medical Images in Frequency Domain using Discrete Curvelet Transform", European Journal of Scientific Research ISSN 1450-216X Vol.60 No.2, pp. 305-315 (2011)

[4] Meneka, R., Chellamuthu, C., "3D Cranio-Facial Surface Modeling from 2D CT Slices Using Fast Corner Detector", Int. Journal of Computer and Electrical Engineering, Vol. 2, No. 5, pp. 1793-8163 (2010).

[5] Janaszewski, M., Couprie, M., Babout, L., "Hole Filling in 3D Volumetric Objects", Pattern Recognition 43, pp. 3548-3559 (2010).

[6] Bagci, E.,"Reverse engineering applications for recovery of broken or worn parts and re-manufacturing: Three case studies", Adv. in Engineering Software 40, pp. 407-418 (2009).

[7] Bi, Z. M., "Advances in 3D Data Acquisition and Processing for Industrial Applications", Robotics and Computer-Integrated Manufacturing, 26, pp. 403-413(2010).

[8] Archip, N., Rohling, R., Dessenne, V., Erard, P. J., Nolte, L. P., "Anatomical Structure Modeling from Medical Images", Computer Methods and Programs in Biomedicine 82 pp. 203-215 (2006).

[9] Ma, Z., Tavares, J. M. R. S., Jorge, R. M. N., "A review of current segmentation techniques for Medical Images", IMAGAPP 2009 - 1stInt. Conf. on Computer Imaging Theory and Applications, Lisboa, Portugal, Feb. 5-8, 2009.

[10] Sun, W., Starly, B., Nam, J., Darling, A., "Bio-CAD modeling and its applications in computer-aided tissue

engineering", Computer-Aided Design 37 pp. 1097-1114 (2005)

[11] Hongsheng, F., Liang, G.,Luyue, J.,Yuzhang, L., "Personalized Cranium Defects Restoration Technique Based on Reverse Engineering", Tsinghua Science and Technology, 14, pp. 82-88 (2009).

[12] Yoo, D. J., "Three-dimensional surface reconstruction of human bone using a B-spline based interpolation approach", Computer-Aided Design 43, pp. 934-947(2011).

[13] Wang, J., Ye, M., Liu, Z., Wang, C.,"Precision of Cortical Bone Reconstruction Based on 3D CT Scans", Computerized Medical Imaging and Graphics, 33, pp. 235-241(2009).

[14] Pereira, N. F.,Sitek, A., "Evaluation of a 3D Point Cloud Tetrahedral Tomographic Reconstruction Method", Phys Med Biol., 55(18), pp. 5341-5361 (2010).

[15] Silva, I. N. L., Barbosa, G. F., Borowski, R., Soares, G., Beltrao, M. C. G., "Creating Three-Dimensional Tooth Models from Tomographic Images", Baltic Dental and Maxillofacial Journal, 10, pp. 67-71 (2008).

[16] Sezgin, M., Sankur, B., "Survey overImage Thresholding Techniques and Quantitative Performance Evaluation", Journal ofElectronic Imaging 13(1), pp. 146-165 (2004).

[17] Zaidi, H., Erwin, W. D., "Quantitative Analysis in Nuclear Medicine Imaging", doi: 10.2967/jnumed.107.042598, J.Nucl.Med.vol. 48(8), 2007.

[18] Zhang, J.,Yan, C. H., Chui, C. K., Ong,S. H., "Fast segmentation of bone in CT images using 3D adaptive thresholding", Computers in Biology and Medicine, 40, pp. 231-236, 2010.

[19] Lagravere, M. O., Fang, Y., Carey, J., Toogood, R. W., "Density conversion factor determined using a cone-beam computed tomography unit NewTom QR-DVT 9000", Dentomaxillofacial Radiology, 35(6), pp. 407-409, 2006.

[20] Grove, O., Rajab, K.,Piegl, L. A., Lai-Yuen, S., "From CT to NURBS: Contour Fitting with B-spline Curves", Computer-Aided Design & Applications, 8(1), pp. 3-21, 2011.

[21] Stanberry, L., Besag, J., "Boundary reconstruction in binary images using splines, Pattern Recognition", Pattern Recognition, Vol. 47, Issue 2, pp. 634-642, 2014.