3D reconstruction of long bones utilising magnetic resonance imaging (MRI)

Magnetic resonance imaging MRI has been shown to be a potential alternative in the previous studies conducted using small bones tarsal bones and parts of the long bones.. Furthermore, so

Utilising Magnetic Resonance Imaging

This thesis is submitted in fulfilment of the requirements for the

degree of Doctor of Philosophy

Institute of Health and Biomedical Innovation

School of Engineering Systems Faculty of Built Environment and Engineering

Queensland University of Technology

Brisbane, Australia

2011

Abstract

The design of pre-contoured fracture fixation implants (plates and nails) that correctly fit the anatomy of a patient utilises 3D models of long bones with accurate geometric representation 3D data is usually available from computed tomography (CT) scans of human cadavers that generally represent the above 60 year old age group Thus, despite the fact that half of the seriously injured population comes from the 30 year age group and below, virtually no data exists from these younger age groups to inform the design of implants that optimally fit patients from these groups Hence, relevant bone data from these age groups is required The current gold standard for acquiring such data–CT–involves ionising radiation and cannot be used

to scan healthy human volunteers Magnetic resonance imaging (MRI) has been shown to be a potential alternative in the previous studies conducted using small bones (tarsal bones) and parts of the long bones However, in order to use MRI effectively for 3D reconstruction of human long bones, further validations using long bones and appropriate reference standards are required

Accurate reconstruction of 3D models from CT or MRI data sets requires an accurate image segmentation method Currently available sophisticated segmentation methods involve complex programming and mathematics that researchers are not trained to perform Therefore, an accurate but relatively simple segmentation method is required for segmentation of CT and MRI data Furthermore, some of the limitations

of 1.5T MRI such as very long scanning times and poor contrast in articular regions can potentially be reduced by using higher field 3T MRI imaging However, a quantification of the signal to noise ratio (SNR) gain at the bone - soft tissue interface should be performed; this is not reported in the literature As MRI scanning

of long bones has very long scanning times, the acquired images are more prone to motion artefacts due to random movements of the subject‟s limbs One of the artefacts observed is the step artefact that is believed to occur from the random movements of the volunteer during a scan This needs to be corrected before the models can be used for implant design

As the first aim, this study investigated two segmentation methods: intensity thresholding and Canny edge detection as accurate but simple segmentation methods for segmentation of MRI and CT data The second aim was to investigate the

usability of MRI as a radiation free imaging alternative to CT for reconstruction of 3D models of long bones The third aim was to use 3T MRI to improve the poor contrast in articular regions and long scanning times of current MRI The fourth and final aim was to minimise the step artefact using 3D modelling techniques

The segmentation methods were investigated using CT scans of five ovine femora The single level thresholding was performed using a visually selected threshold level

to segment the complete femur For multilevel thresholding, multiple threshold levels calculated from the threshold selection method were used for the proximal, diaphyseal and distal regions of the femur Canny edge detection was used by delineating the outer and inner contour of 2D images and then combining them to generate the 3D model Models generated from these methods were compared to the reference standard generated using the mechanical contact scans of the denuded bone The second aim was achieved using CT and MRI scans of five ovine femora and segmenting them using the multilevel threshold method A surface geometric comparison was conducted between CT based, MRI based and reference models To quantitatively compare the 1.5T images to the 3T MRI images, the right lower limbs

of five healthy volunteers were scanned using scanners from the same manufacturer The images obtained using the identical protocols were compared by means of SNR and contrast to noise ratio (CNR) of muscle, bone marrow and bone In order to correct the step artefact in the final 3D models, the step was simulated in five ovine femora scanned with a 3T MRI scanner The step was corrected using the iterative closest point (ICP) algorithm based aligning method

The present study demonstrated that the multi-threshold approach in combination with the threshold selection method can generate 3D models from long bones with an average deviation of 0.18 mm The same was 0.24 mm of the single threshold method There was a significant statistical difference between the accuracy of models generated by the two methods In comparison, the Canny edge detection method generated average deviation of 0.20 mm MRI based models exhibited 0.23 mm average deviation in comparison to the 0.18 mm average deviation of CT based models The differences were not statistically significant 3T MRI improved the contrast in the bone–muscle interfaces of most anatomical regions of femora and tibiae, potentially improving the inaccuracies conferred by poor contrast of the articular regions Using the robust ICP algorithm to align the 3D surfaces, the step

artefact that occurred by the volunteer moving the leg was corrected, generating errors of 0.32 ± 0.02 mm when compared with the reference standard

The study concludes that magnetic resonance imaging, together with simple multilevel thresholding segmentation, is able to produce 3D models of long bones with accurate geometric representations The method is, therefore, a potential alternative to the current gold standard CT imaging

Contents

Abstract III

Keywords VI

List of figures XIII

List of tables XV

Publications, presentations and awards XVI

Authorship XIX

Acknowledgement XXI

Abbreviations XXII

Chapter 1 Introduction 1

Chapter 2 Quantitative imaging of the skeletal system for 3D reconstruction (Background) 7

2.1 Introduction 7

2.2 Computed tomography (CT) 8

2.2.1 Basic principles of CT 8

2.2.2 Radiation exposure during CT imaging 9

2.3 Magnetic resonance imaging (MRI) 10

2.3.1 Basic principles of MRI 10

2.3.2 How tissue contrast is determined 12

2.3.3 Selection of slice position and thickness 13

2.3.4 Pulse sequences 14

2.3.5 MRI safety 14

2.3.6 Signal to noise ratio of an MRI system 14

2.3.7 Artefacts of MRI 15

2.3.7.1 Motion artefacts 16

2.3.7.2 Magnetic susceptibility difference artefact 16

2.3.7.3 Chemical shift 17

2.3.8 MRI for imaging of the skeletal system 17

2.3.9 Advantages and current limitations of MRI 18

2.3.9.1 Longer scanning times of MRI 18

2.3.9.2 Poor contrast in certain anatomical regions 18

2.3.9.3 Non-uniformity of the external magnetic field 19

2.3.9.4 Limited accessibility 19

2.4 Summary 20

Chapter 3 Image processing and surface reconstruction 21

3.1 Introduction 21

3.2 Acquisition of data for 3D modelling of bones 22

3.2.1 Effect of in plane resolution and slice thickness on accuracy of reconstructed 3D models 23

3.3 Image segmentation 24

3.3.1 Manual segmentation 25

3.3.2 Intensity thresholding 25

3.3.2.1 Selecting an appropriate threshold level 26

3.3.2.2 Multilevel thresholding 26

3.3.3 Edge detection 28

3.3.4 Region growing 28

3.3.5 Sophisticated segmentation methods 29

3.4 Surface generation 29

3.5 Registration (aligning) and comparison of surfaces 30

3.6 A reference standard for validating 3D models of bones 30

3.7 Aims of the study 32

3.8 Methods 32

3.8.1 Samples 32

3.8.2 Image segmentation 32

3.8.3 Reference model for validation of the outer 3D models 33

3.8.3.1 Removal of the soft tissues from long bones 33

3.8.3.2 Scanning of the bone‟s outer surface using the contact scanner 34

3.8.3.3 Reconstruction of the 3D model from scanned surfaces 37

3.8.4 Reference model for validation of the medullary canal 39

3.8.5 Basic 3D modelling techniques using Rapidform 2006 41

3.8.5.1 Registration of 3D surfaces using Rapidform 2006 41

3.8.5.2 Comparison of the aligned 3D models 43

3.8.5.3 Dividing the 3D models of bones into anatomical regions 44

3.9 Results 44

3.10 Summary, discussion and conclusion 45

3.11 Paper 1: Effect of CT image segmentation methods on the accuracy of long bone 3D reconstructions (published) 48

Chapter 4 Application of 3D modelling techniques for orthopaedic implant design and validation 57

4.1 Introduction 57

4.2 3D models for implant design and validation 58

4.3 Aims of the study 59

4.4 Methods 59

4.5 Results 59

4.6 Summary, discussion and conclusion 60

4.7 Paper 2: Quantitative fit assessment of tibial nail designs using 3D computer modelling (published) 61

Chapter 5 Magnetic resonance imaging for 3D reconstruction of long bones 67

5.1 Introduction 67

5.2 Imaging of skeletal system with MRI 68

5.3 Aims of the study 71

5.4 Methods 71

5.5 Results 72

5.6 Summary, discussion and conclusion 72

5.7 Paper 3: Quantification of the accuracy of MRI generated 3D models of long bones compared to CT generated 3D models (in press) 74

Chapter 6 Higher field strength MRI scanning of long bones for generation of 3D models 83

6.1 Introduction 83

6.2 Theoretical consideration of increased SNR at 3T 84

6.3 3T MRI for musculoskeletal system imaging 84

6.3.1 Spin relaxation times and flip angle 85

6.3.2 Fat suppression 86

6.3.3 Magnetic susceptibility at 3T MRI 87

6.3.4 Chemical shift at 3T 87

6.3.5 MRI safety at 3T 88

6.4 Aims of the study 88

6.5 Methods 88

6.5.1 Samples 88

6.5.2 Measuring the quality of MR images 88

6.5.3 Quantification of spin relaxation times 90

6.5.4 Comparison of 1.5T and 3T imaging of musculoskeletal system 91

6.6 Results 93

6.7 Summary, discussion and conclusion 93

6.8 Paper 4: 3T MRI improves bone-soft tissue image contrast compared with 1.5T MRI (Submitted – under review) 96

Chapter 7 Step artefact caused by Magnetic Resonance Imaging of long bone 121

7.1 Introduction 121

7.2 Motion artefact of MRI 122

7.3 Aims of the study 123

7.4 Methods 123

7.5 Results 124

7.6 Summary, discussion and conclusion 124

7.7 Paper 5: Correction of step artefact associated with MRI scanning of long bones (Submitted – under review) 126

Chapter 8 Summary, conclusion and future directions 145

8.1 Summary and conclusion 145

8.2 Future directions 148

Appendix 1 Ethical approval for the study in Chapter 6 151

Appendix 2 Participant information and Consent form used in Chapter 6 154

Appendix 3 Animal tissue use notification 157

References 159

List of figures

Figure 2.1: Arrangement of the x-ray source, detector and the object in a CT scanner 9

Figure 2.2: A spin possesses a tiny magnetic field aligned with the axis of rotation 11

Figure 2.3: Spins aligned with the external magnetic field B0 11

Figure 2.4: An MRI image of the coronal section of the proximal femur 19

Figure 2.5: The uniform regions of the external magnetic field of a MRI scanner (The uniform region is shaded) 19

Figure 3.1: Average intensity values of the outer bone contours as detected by the Canny filter for each axial CT image 27

Figure 3.2: The process of removing soft tissues from the sheep femur before scanning with the contact mechanical scanner: a - gross dissection with the scalpel, 34

Figure 3.3: Scanning of the bone's outer surface of the diaphyseal region using the MDX 20 contact scanner (The bone is positioned on the stage using glue tags) 35

Figure 3.4: Bone is cut in three parts in order to scan the articular surfaces which cannot be reached by the scanner on the intact bone 36

Figure 3.5: Positioning of the proximal articular segment of the femur in order to scan the articular surface 37

Figure 3.6: The reconstructed model before the scanning of articular surfaces (This model was used as a guide to scan the articular regions) 37

Figure 3.7: Scanned surface with unusable data 38

Figure 3.8: The surface after removing the unusable data 38

Figure 3.9: Two adjacent surfaces are fine registered 39

Figure 3.10: The final 3D model reconstructed by merging the surfaces 39

Figure 3.11: a - The original microCT image (a cross section from the diaphysis); and b - the image after applying a 20 × 20 median filter 40

Figure 3.12: The initial aligning of the CT based 3D model to the reference model using

Trackball prior to the application of fine registration function 42

Figure 3.13: A CT based model (red) is aligned to the reference model (blue) in

Rapidform 2006 using the fine registration function 42

Figure 3.14: Comparison of the aligned CT model to the reference model in Rapidform

2006 43

Figure 3.15: Five anatomical regions used for the comparison: 1 - femoral head, 2 -

proximal region, 3 - diaphysis, 4 - distal region, 5 - distal articular region 44

Figure 3.16: Reference planes and curves used for the splitting of the model into five

anatomical regions 44

Figure 5.1: Cross sections of CT (left) and MRI (right) from the same anatomical location

of a sample 69

Figure 6.1: Positioning of the volunteer in the MRI scanner and the position of the matrix

coils that cover the lower limbs and the pelvis 92

Figure 6.2: Positioning of the field of view (FOV) on volunteer‟s leg 93

Figure 7.1: The step artefact caused by volunteer moving the leg between two successive

scanning stages 121

Figure 7.2: MRI scanning of human lower limb with five scanning segments to scan the

complete limb 123

List of tables

Table 3.1 Specifications of the MDX 20 contact 3D scanner 35

Table 3.2 Scanner parameters used for microCT scanning 40

Table 6.1 TR and TE values used for the MRI scanning at 1.5T and 3T 90

Table 6.2 Different flip angles used for scanning 90

Table 6.3 The protocols used for MRI scanning 92

Publications, presentations and awards Journal Publications

1 Rathnayaka K, Schuetz MA, Sahama T & Schmutz B Correction of step

artefact associated with MRI scanning of long bones, Submitted to Medical

Engineering and Physics

2. Rathnayaka K, Coulthard A, Momot K, Volp A, Sahama T, Schuetz MA &

Schmutz B 3T MRI improves bone-soft tissue image contrast compared

with 1.5T MRI, submitted to Magnetic Resonance Imaging

3 Rathnayaka K, Momot K I, Noser H, Volp A, Schuetz M, Sahama T &

Schmutz B Quantification of the accuracy of MRI generated 3D models of

long bones compared to CT generated 3D models Medical Engineering &

Physics 2011, in press, DOI:10.1016/j.medengphy.2011.07.027

4 Rathnayaka K, Schmutz B, Sahama T and Schuetz M A Effects of CT image

segmentation methods on the accuracy of long bone 3D reconstructions

Medical Engineering & Physics 2011, 33(2): 226-233

5 Schmutz B, Rathnayaka K, Wullschleger ME, Meek J, Schuetz MA

Quantitative fit assessment of tibial nail designs using 3D computer modeling Injury 2010; 41(2): 216-219

1 Rathnayaka K, Cowin G, Schuetz MA, Sahama T, Schmutz B Correction of

the step artefact in 3D bone models caused by the random movement of the lower limb during MRI 17th Annual Scientific Meeting, Australian & New Zealand Orthopaedic Research Society Brisbane, Australia, 1-2 September,

2011 (Oral presentation)

2 Rathnayaka K, Coulthard A, Momot K, Volp A, Sahama T, Schuetz M,

Schmutz B Improved image contrast of the bone-muscle interface with 3T

Singapore, 1-7 August, 2010 (Poster presentation)

3 Schmutz B, Rathnayaka K, Wullschleger M, Meek J, Schuetz M

Quantitative fit assessment of tibial nail designs using 3 D computer modeling German Society for Orthopaedic and Trauma Surgery Berlin,

Germany, 21-24 October, 2009 (Oral Presentation)

1

The conference abstracts have not been included in the thesis as the contents of them are covered by the journal articles

4 Rathnayaka K, Sahama T, Schuetz MA, Schmutz B Validation of 3D models

of the outer and inner surfaces of an ovine femur 15th Annual Scientific Meeting, Australian & New Zealand Orthopaedic Research Society Adelaide, Australia, 9-10 October, 2009 (Oral presentation)

5 Rathnayaka K, Momot K, Volp A, Noser H, Sahama T, Schuetz MA,

Schmutz B Quantification of the accuracy of MRI generated 3D models of

long bones 4th Asian Pacific conference of biomechanics University of Canterbury, Christchurch, New Zealand, 14–17 April, 2009 (Oral presentation)

6 Rathnayaka K, Schmutz B, Sahama T, Schuetz MA Effects of image

segmentation methods on the accuracy of long bone 3D reconstructions

14th Annual Scientific Meeting, Australian & New Zealand Orthopaedic Research Society Brisbane, Australia, 17-18 November, 2008 (Poster presentation)

7 Mohd Radizi S, Rathnayaka K, Pratap J, Mishra S, Schuetz MA, Schmutz B,

The effects of CT convolution kernels on the geometry of 3D bone models

14th Annual Scientific Meeting, Australian & New Zealand Orthopaedic Research Society Brisbane, Australia, 17-18 November, 2008 (Poster

presentation)

Awards and Scholarships

1 Outstanding HDR student of the month, Faculty of Built Environment and Engineering, Queensland University of Technology, December 2010

2 Student travel grant awarded by 6th World congress on Biomechanics

Singapore, 1-7 August, 2010

3 Joint winner of the Wilhelm-Roux-Preis 2009, at Annual conference of the German Society for Orthopaedic and Trauma Surgery Berlin, Germany, 21-

24 October, 2009

4 Student Travel grant awarded by 15th Annual Scientific Meeting of Australian

& New Zealand Orthopaedic Research Society Adelaide, Australia, 9-10 October, 2009

5 Runner-up for best poster presentation, IHBI inspires postgraduate student conference Gold Coast, Australia, 2-4 December, 2008

6 QUT, Faculty of Built Environment and Engineering living allowance PhD scholarship 2008-2011

………

Rathnayaka Mudiyanselage

Acknowledgement

Firstly, I offer my sincere thanks to my supervisors: Dr Beat Schmutz for the invaluable support, guidance and advice given throughout the PhD and for helping to establish my directions; Prof Michael Schuetz, my principal supervisor, for encouragement and guidance given; and Dr Tony Sahama for introducing me to the trauma research group at Queensland University of Technology and for the support given throughout the PhD study

I offer my special thanks also to: Dr Konstantin Momot for helping me by sharing his knowledge of MRI physics and by reading manuscripts, especially during the second and fourth parts of the research project; Prof Alan Coulthard for collaborating with

me for the fourth part of the project; Dr Gary Cowin for helping me with MRI scanning for the third part of the project; Mr Andrew Volp and Mr Russell Porter at Princes Alexandra Hospital; Mr Raymond Buckley at Royal Brisbane and Women‟s Hospital for MRI scanning of the samples and volunteers of the study; Mr Jit Pratap

at Princes Alexandra Hospital; and Ms Margaret Day at University of Queensland for

CT scanning of samples

I must offer my sincere gratitude to all those who volunteered as subjects for the study and spent their valuable time on my project, and to the National Imaging Facility for providing me with 100% subsidised access to the 3T MRI scanner Thanks to all the researchers who donated ovine limbs from their studies and who helped me to obtain them at the Medical Engineering Research Facility (MERF) I would also like to thank the High Performance Computer (HPC) Unit and its personnel at QUT for their help with the 3D modelling and use of the super computers Thanks to all the members of the trauma research group and all the friends who helped me with various aspects of this research, especially with feedback

on writing and presentations Finally, the laboratory and directorate staff at IHBI and MERF also provided kind help during this project and I offer them my gratitude

CAS Computer assisted surgery

CNR Contrast to noise ratio

ICP Iterative closest point

M0 Net magnetisation vector

MHz Mega Hertz

MR Magnetic resonance

MRI Magnetic resonance imaging

Mt Transverse component of net magnetisation vector

Mz Longitudinal component of net magnetisation vector

NAV Number of signal averages

NMR Nuclear magnetic resonance

NPA Number of acquired partitions

NPE Number of acquired phase encodes

PMMA Poly-methyl methacrylate (Dental acrylic)

RF Radiofrequency

ROI Region of interest

SD Standard deviation

SNR Signal to noise ratio

SNRGER Signal to noise ratio for gradient echo sequence

SNRSE Signal to noise ratio for spin echo sequence

T1 Longitudinal relaxation time

T2 Transverse relaxation time

TE Echo time

TMS Tetramethylsilane

TR Repetition time

V Voxel volume

Chapter 1

Introduction

The introduction of x-ray computed tomographic (CT) scanning and magnetic resonance imaging (MRI) in the 1970s allowed medical personnel and researchers to visualise the internal anatomical structures of the human body in three dimensions This allowed clinicians and researchers to reconstruct anatomical structures as computer based three dimensional (3D) models and perform various experiments that cannot be performed on living subjects Thus, accurate reconstruction of 3D models

of anatomical structures from CT and MRI became a major research interest Even though the main mode of imaging bones is CT, the involvement of ionising radiation leads clinicians and researchers to avoid CT whenever possible Thus, a trend towards the frequent use of MRI is developing among these groups, not only due to the non-involvement of ionising radiation in MRI, but also due to its ability to provide better quality images of soft tissue

Reconstruction of a three dimensional computer model of an anatomical structure using either CT or MRI imaging methods involves a number of complex processes: data acquisition; segmentation of the region of interest (ROI) and surface generation from the segmented volume Each of these processes plays a crucial role in determining the geometric accuracy of the reconstructed 3D model Since the geometric accuracy of 3D models is of high importance for most of their applications (e g implant design and simulation of surgery), these processes in reconstructing 3D models have drawn major attention from researchers [1-3] While all steps play a crucial role in determining the accuracy of 3D models, image segmentation is one of the steps which has a higher human involvement and is thus vulnerable to errors Even though existing sophisticated segmentation methods are capable of minimising the human intervention, most of these methods involve complex programming and mathematics which many of the researchers are not trained to perform [2, 4-7]

Furthermore, these algorithms are designed to perform segmentation in a specific anatomical region and, therefore, are not easily extended to the segmentation of a different region due to their complex nature Thus, a simple but accurate method for medical image segmentation is a necessity

Reconstruction of a 3D model of a small bone (phalanges or metatarsal bones) is relatively easy when compared to the reconstruction of a 3D model of a long bone that has a complex geometry Thus, most of the studies that investigated segmentation methods have utilised small bones Nevertheless, 3D reconstruction of long bones is important as most of the fracture fixation plates and intramedullary nails are used for fixation of long bones When 3D models of long bones are reconstructed, the diaphyseal as well as the distal and proximal regions are equally important Most of the fracture fixation plates and intramedullary nails extend to the proximal and distal regions (e.g expert tibia nail used in chapter 4) The intramedullary nail insertion point is usually in the proximal or distal region, thus, accuracy of these regions are important to determine the entry point of the nail Furthermore, design of implants such as joint replacements needs highly accurate 3D models of the proximal and distal articular regions Therefore, the research projects contained in the thesis will focus on all anatomical regions of long bones

The decision to use either CT or MRI is mainly determined by the anatomical structures being scanned While CT visualises the bone tissue with better contrast, MRI visualises soft tissues with better contrast as its main source of signal is hydrogen nuclei which are abundant in soft tissues The radiation exposure of CT limits its utilisation to clinical cases and cadaver specimens As most of the available cadavers are more than 60 years old, the data acquisition is also limited to this age group However, approximately 51% of land transport trauma patients (or 11.4% of total injury hospitalisations) in Australia during the 2006-2007 period were under 30 years of age Furthermore, the study conducted by Noble et al 1995 [8] shows that the femoral isthmus expands in old female population compared to the young population The study also showed that the medullary canal expands and the cortex becomes thinner in old females and the CCD angle (femoral neck-shaft angle) change with the age These changes will impact the anatomical fitting of plates and intramedullary nails designed using 3D models reconstructed from old bones In addition, osteophytes in old bones can significantly affect the anatomical fitting of

fracture fixation plates especially in ends of the bones Thus, the acquisition of bone data from this age group to inform the design of anatomically shaped fracture fixation implants (plates and nails) for its trauma patients is of utmost importance [9] As MRI does not utilise ionising radiation, it is a potential alternative to CT for acquiring bone data of volunteers from younger age groups

Even though MRI visualises soft tissues with a high contrast, due to the extremely short transverse relaxation times, bones generally do not generate a signal in MRI [10-12] However, using the signal generated by the soft tissues, bone geometry can

be delineated from the surrounding soft tissues and this has been demonstrated in the literature [1, 13-17] MRI has been used for the scanning of bones mainly in the case

of diagnosing metastatic disease, as MRI visualises metastasis with better quality [18]

The use of MRI for 3D reconstruction of bones has been reported in computer assisted surgery (CAS) and in foot bone motion quantification where the 3D models

of vertebrae and tarsal bones have been reconstructed [19-21] Most of these studies have used MRI for small bones with relatively simple geometry, and a proper validation of the models has not been performed Lee et al used MRI to generate a 3D model of a porcine femur; however, the model has not been validated using an accurate validation standard [1] Therefore, before using MRI for 3D reconstruction

of long bones, a quantitative validation with an accurate reference standard is necessary

Some of the current limitations of the MRI scanning of long bones are long scanning times and the difficulty of segmenting certain anatomical regions, conferred by poor contrast between those anatomical regions and surrounding soft tissues Since the signal to noise ratio (SNR) of an MRI system is approximately directly proportional

to the main magnetic field of the scanner, higher field strength (3T) scanners promise

to offer an improved signal which can be converted to faster scanning times or better image quality compared to the currently available (1.5T) scanners [22, 23] The improved image quality of 3T scanners has been demonstrated in a few studies for computer assisted surgery and kinematic analysis of foot bone motion [24, 25] However, the contrast at the bone muscle interface, which is more important for segmentation of bones, has not been quantified and compared in those studies

Furthermore, different contrast levels which occur in different anatomical regions of

a long bone need to be studied in detail to see the improvement in contrast at those regions

MR imaging of anatomical structures is challenged by various artefacts Among them, the motion artefacts due to random movements are of main concern in MRI imaging of long bones due to their effects on the geometric accuracy of 3D models reconstructed In addition to the long scanning time of MRI, the non-uniformity of the main magnetic field limits the effective scanning length, resulting in a long bone being scanned in several stages One of the adverse effects of this, which has been observed in an initial study conducted by the supervisory team, is the displacement artefact caused by the volunteer moving the leg between two scanning stages [26] Thus, there is a step that can be seen on the final 3D model generated from such a data set This artefact may not be critical for clinical use of the images; however, when the precise measurements are performed for implant design, these artefacts can have a major effect on their accuracy Therefore, minimisation or correction of these artefacts can improve the accuracy of implants designed using those models

This thesis presents the studies carried out to investigate: a simple and accurate method for medical image segmentation; the feasibility of MRI as an alternative to

CT for scanning of long bones; the usability of higher field strength MRI to overcome some of the problems with low field strength scanners; and the correction

of the step artefact that occurred from MRI scanning of long bones Chapter 2 provides the basic physics involved in CT and MRI, while Chapter 3 provides the background of image segmentation, 3D reconstruction and the investigation carried out to develop and validate a simple and accurate image segmentation method Chapter 4 presents the application of 3D modelling techniques in implant validation, utilising 3D models of long bones for fit quantification of two anatomically shaped intramedullary nails Chapter 5 presents the investigation carried out to formally validate the MRI based 3D models of long bones against the CT based models Chapter 6 provides the details of the quantitative comparison between 1.5T MRI and 3T MRI Chapter 7 presents the correction of the step artefact that occurred due to the random movements of the volunteer during MRI scanning Chapter 8 presents a summary, discussion and future directions of the thesis

The aims of the study in brief are as follows:

Investigation of the accuracy of multilevel intensity thresholding and Canny edge detection for segmentation of CT images

Quantification of the accuracy of 3D models based on MRI compared to the 3D models based on CT

Quantitative comparison of the image quality at 1.5T MRI to 3T MRI

Correction of the step artefact that occurs due to the random movement of the lower limb during MR imaging

Chapter 2

Quantitative imaging of the skeletal

system for 3D reconstruction

(Background)

A number of methods are available for the imaging of various anatomical structures

of the human body, such as: plain ray, computed tomography (CT), Dual energy ray absorptiometry (DEXA), magnetic resonance imaging (MRI) and ultrasound (US) Even though quantitative imaging of the skeletal system is possible with most

x-of the above scanning methods, accurate spatially-resolved information x-of the anatomical structures can only be acquired using CT or MRI Thus, CT and MR imaging methods have taken an integral part in research and in clinical applications where the 3D reconstruction of the anatomical structures is required The most commonly used imaging technique for quantitative imaging of the skeletal system is CT; however, MRI has also been reported as a potential imaging technique for this purpose

CT has become the gold standard of imaging the skeletal system for 3D reconstruction because CT produces images with better contrast at the bone–soft tissue interface CT images can also be acquired within a very short period of time, thus, essentially avoiding the motion artefacts caused by moving body parts or tissues While CT involves ionising radiation that prevents its use on healthy human volunteers for research purposes, it can be used for in vitro research studies for scanning of bones In the present study, CT will be used to validate two image segmentation methods and for validation of MRI based models of ovine femora Section 2.2 of this chapter provides the basic principles of CT and discusses its advantages and disadvantages

MRI utilises the principles of nuclear magnetic resonance (NMR) of hydrogen nuclei

to generate a signal from the tissue Even though the bone tissues do not generate a significantly large signal, by utilising the signal generated from the surrounding soft tissues, MRI can be used in the 3D reconstruction of bones Thus, MRI can potentially be used to image healthy human volunteers for research purposes, without having to expose them to the ionising radiation of CT Section 2.3 discusses the basic principles, advantages and current problems of MRI in detail Since MRI has been utilised for most of the studies presented in this thesis, it will be discussed in more detail than CT

Computed tomography (CT) was the first method of imaging anatomical structures inside the body without having the problem of the superimposition of anatomical structures that was a major drawback of plain X-ray images Since its introduction to clinical use in 1970 [27], CT has become the most commonly used imaging technique in the clinical setting It has also become the standard practice for imaging

of trauma patients for accurate diagnosis of bone fractures in emergency situations [28, 29]

CT images are acquired by recording the object‟s attenuation of the radiation which

is emitted from an x-ray source (x-ray tube) A CT image is reconstructed from a large number of projections of the object, taken around a single axis of rotation using

an x-ray beam Depending on its x-ray absorption properties, when the x-ray beam passes through the object, a projected image is generated on the detector (image sensor) These images are integrated using a computer based algorithm to produce axial image slices The projections are obtained by rotating the detector and the x-ray

source simultaneously around the object (Figure 2.1)

Figure 2.1: Arrangement of the x-ray source, detector and the object in a basic CT

scanner (A large number of projections of the object will be obtained by rotating the source and the detector simultaneously around the object.)

Early generation CT scanners imaged a patient slice by slice with specific slice spacing Once an image slice is obtained, the table with the patient moves a set distance and the next slice is obtained With the development of helical CT, continuous imaging is performed by moving the patient continuously through the gantry in combination with the continuous rotation of the x-ray source and detector system The obtained data volume is later reconstructed to image slices with specific slice spacing This also allows for the reconstruction of images in anatomical planes other than the traditional axial image slices Modern spiral scanners with multiple rows of x-ray detectors (multi-slice scanners or multi-row scanners) can image a subject within a very short time period (a few seconds), thus almost eliminating motion artefacts

Due to the high accuracy obtained for the bone geometry, CT has become the gold standard for imaging of the bones for reconstructing 3D models, mainly for the development of implants and clinical applications CT can also be used for measurement of relative tissue density and can be presented as Hounsfield Units (HU) for comparison with other or reference tissues

The use of diagnostic CT has increased dramatically over the last 20 years and it is the gold standard for bone imaging However, CT uses a high dose of radiation and concerns have been raised regarding cumulative radiation exposure and associated lifetime risk as there is epidemiological evidence of a small risk of radiation associated cancer at doses comparable to a few CT scans [30-33] For example, the radiation exposure of a standard thoracic CT is equivalent to 400 standard chest x-ray

radiographs (8mSv), and that of a pelvic CT is equivalent to 250 chest radiographs (250 mSv) [30, 34, 35] According to a report by the Royal College of Radiologists

in the UK, CT scans probably contribute almost half of the collective dose of radiation from all x-ray examinations [36] This has become a major problem, as CT scanning of a healthy human volunteer for research purposes is ethically not justifiable due to this high radiation exposure

As a solution, protocols that use low radiation doses while maintaining a higher image quality are under investigation [37-39] Some slice selection strategies (e.g use of fewer slices for simple geometric shapes such as diaphyseal region) have also been investigated to reduce the radiation dose [38] However, due to the fact that the radiation exposure of CT cannot be eliminated completely, some countries do not approve the scanning of volunteers with these protocols Therefore, researchers are moving towards using an imaging technique such as MRI that does not utilise ionising radiation

Magnetic resonance imaging (MRI) utilises the nuclear magnetic resonance (NMR)

of 1H nuclei as the source of signal There are a number of elements that demonstrate NMR capabilities, such as 1H, 13C, 15N, 31P Human tissue is largely composed of water (H2O) and, thus, 1H is the most abundant NMR capable nuclei in the human tissue Throughout this discussion, 1H nuclei are also referred to as „spins‟, as 1

H nuclei have the quantum mechanical property termed „nuclear spin‟

If a single 1H nucleus is considered, it possesses a magnetic moment, which is a

quantum mechanical property, parallel to its axis (Figure 2.2) In the absence of an

external magnetic field, the axes of the spins are randomly aligned in a given tissue

sample and the vector sum of the magnetisation is equal to zero (Figure 2.2)

Figure 2.2: A spin possesses a tiny magnetic field aligned with the axis of rotation

(left); randomly aligned axes of spins in the absence of an external magnetic field (right) [40]2

To measure NMR of 1H nuclei (or any NMR capable nucleus), an external magnetic field (also referred to as „the main magnetic field‟ or „B0‟) is applied to the sample, thus making randomly aligned spins partially align with the externally applied magnetic field (in the opposite direction to B0) (Figure 2.3) Thus, the sample now

possesses a net magnetisation vector (M0) parallel to B0 M0 can be split into its component vectors: Mz which is parallel to B0, and Mt which is perpendicular to B0

At rest, Mz = M0 and Mt = 0 (Figure 2.3)

Figure 2.3: Spins aligned with the external magnetic field B0 and M0 and its two components, Mz and Mt

2 Adapted from: Brown, M.A and Semelka, R.C MRI Basic principles and applications, 4th ed 2010, New Jersey: John Wiley

& Sons

The frequency at which the spins precess under an external magnetic field is proportional to the strength of the external magnetic field and is expressed by the

Larmor equation (Equation 2.1):

of the spins towards a direction perpendicular to B0 (if a 90º pulse is applied), generating a net transverse magnetisation (Mt), and leading Mz to decline At this stage, the Larmor precession of the spins will induce a voltage in the receiving coil (RF coil) which is measured as the MR signal The intensity of the signal generated

in the receiver coil is proportional to the transverse magnetisation (Mt); therefore, the initial magnitude of the signal depends on the value of the Mz immediately prior to the RF pulse

When excited, the angle at which M0 is oriented relative to B0 is the flip angle (FA), which is one of the parameters that should be changed accordingly to get an optimal contrast When the RF wave is shut off, Mz starts to recover, and the inverse of the rate constant of recovery is called the „longitudinal relaxation time‟ (T1) At the same time, Mt starts to decay and the exponential rate constant of decay is called

„transverse relaxation time‟ (T2) Both T1 and T2 take different values for different tissue types [41]

In MRI, tissue contrast is related to the differences of rate of magnetisation decay The three factors that determined the tissue contrast in the present work were T1, T2

and the proton density of the tissue The differences between spin relaxation times and the proton density in different tissues serve as the basis for image contrast The contrast can be manipulated by selecting different scan parameters, namely repetition time (TR) and echo time (TE) TR is the time period between two successive

acquisitions and TE is the time between delivery of the RF pulse and signal detection T1 contrast can be selected by choosing short repetition times (TR) and such images are called „T1 weighted images‟ where the contrast is mainly determined

by T1 of the particular tissue T2 contrast can be modulated by changing echo time (TE) and the images of which the contrast is mainly determined by T2 are called „T2

weighted images‟ In both types of images, there is a contribution from T1 and T2, however, the effect from one is minimised while the other is maximised The contrast can also be determined by the proton density of the tissue and the images acquired this way are called „proton density weighted images‟

2.3.3 Selection of slice position and thickness

The slice position, slice thickness and the Phase and Read directions are determined

by the respective gradient pulses and (in the case of the slice position) the RF frequency offset When a magnetic field gradient is applied on top of the existing main field B0 in x, y, or z directions, the spins at different locations along the gradient experience slightly different magnetic fields Thus, the spins at different locations along the gradient precess at different Larmor frequencies, which are given

by the following equation (2.2):

Three linear mutually perpendicular gradients are used: phase encoding gradient, readout gradient and slice selection gradient The phase-encoding gradient encodes the locations of the nuclei in the direction of that gradient using the phase accumulated by the nuclei during the gradient pulse The readout (or frequency-encoding) gradient encodes the locations of the nuclei in the direction of that gradient using the position-dependent precession frequency during acquisition of the echo The receiver coils detect the entire spectrum of the different precession frequencies during the readout gradient, which ensures that the field of view in the Read direction covers the entire sample

The slice selection gradient is used to achieve the localisation of the RF excitation to

a region in the space The RF pulse applied has two parts: a central frequency and a narrow bandwidth of frequencies (1-2 kHz) When such RF pulse is applied to the sample in the presence of the slice selection gradient, a narrow region of tissue achieves the resonance state Thus the bandwidth of the applied RF wave determines the thickness of the image slice

The pulse sequence is a sequence of instructions to the hardware for switching the

RF pulse and gradient pulses on and off and for sampling the signal, keeping a specific time period between each of them This allows for the acquisition of data in the desired manner by manipulating the relevant parameters (TR, TE, and FA) Spin echo sequence and gradient echo sequence are two commonly used sequences for clinical imaging The FLASH (Fast Low Angle Shot) sequence used for this study is based on the gradient echo sequence

MRI is relatively safe compared to CT; however, the RF power deposition in the conductive tissue results in heating of the tissue inside the body To prevent hazards from the heat, the specific absorption rates (SAR) of energy dissipation are monitored using hardware level or software level monitors [42] There are no known direct biological hazards to patients from exposure to strong magnetic fields However, there is a high risk of the strong magnetic field of the scanner affecting metallic implants and cardiac pacemakers Thus, MRI is contraindicated for patients with cardiac pacemakers, metallic debris in eyes or other ferromagnetic materials in the body Patients with implants that do not have a risk of detaching, or which do not contain ferromagnetic materials (e.g hip replacements, stents made of nickel-titanium alloy) can be safely scanned with MRI [27]

Signal to noise ratio (SNR) is an important measure that can be used to quantify the

quality of a MRI system (Equation 2.3) In the case of conducting tissues, the

intrinsic SNR of a MRI system is approximately proportional to the strength of the external magnetic field and the volume of tissue being scanned, and depends on tissue parameters (e.g T1 & T2) The following equations (2.4 & 2.5) show the

relationship between SNR and other parameters of a MRI system for a spin echo sequence and gradient echo sequence [43]:

(

0

T TE T TR AV

PA PE

BW

NNNVB

2 1

1

cos)1

(

)1

(sin

0

T TE T

TR

T TR AV

PA PE

e

eBW

NNNVB

Where SNRSE = signal to noise ratio for spin echo sequence, SNRGER = signal to noise ratio for a gradient echo sequence (FLASH), B0 = external magnetic field, V = voxel volume, NPE = number of acquired phase encode lines, NPA = number of acquired partitions, NAV = number of signals averaged, BW = receiver bandwidth per pixel, TR

= repetition time, T1 = longitudinal relaxation time, TE = echo time, T2 = transverse relaxation time and = flip angle

In both equations, the term under the square root is the total time for acquiring data Therefore, intrinsic SNR is directly proportional to the strength of the external magnetic field, the voxel volume, the square root of total sampling time and contrast related parameters Thus, from the above relationship, it is clear that the external magnetic field, voxel size, number of averages, flip angle, T1, T2, TR and TE all have

an influence on SNR of a MRI system In addition, sensitivity to magnetic susceptibility and chemical shift difference between fat and water also influence the SNR of a MRI system

When the pixels in the MR image do not represent the actual anatomical structure being scanned, this region of the image is referred to as an „artefact‟ These artefacts appear among the general structures as signals that do not correspond to the actual tissue at the location They may or may not be easily recognised from the normal anatomy, particularly if they are low in intensity

2.3.7.1 Motion artefacts

Motion (also referred to as movement) artefacts occur as a result of movement of the tissue (heart, lung) or parts of the body (limbs) which are being scanned during the data acquisition Motion artefacts can either be due to periodic movements (e.g blood flow, respiration and heart beat) or random movements which mainly occur due to the person‟s inability to keep the body parts still during the long scanning time These movements result in misregistration of pixels along the phase-encoding direction [40, 44] The artefact occurs by tissue that is excited at one location producing signals that are mapped to a different location during detection [40] The nature and the extent of the artefact depend on the extent of movement and the protocol used for scanning

The most common motion artefact caused by periodic movements is due to blood flow in the vessels of the tissue being scanned [40] If the blood flow is in a direction perpendicular to the slice plane, the artefact is localised to the vessel diameter If the flow is along the slice plane, a more diffuse artefact is seen

Motion artefacts from random movements occur due to muscle contraction from nerve excitations They can also occur as a result of the volunteer or patient randomly moving the body part being scanned due to the longer scanning times (e.g keeping a lower limb still for 65 minutes is nearly impossible) Since the complete lower limb has to be scanned in several segments, volunteers tend to move the leg between segments and this causes a step in the final image stack

Motion artefacts that occur due to periodic movements such as breathing movements can be minimised by using specially designed protocols which synchronise the data acquisition with the breathing movements, or by post processing techniques Elimination of the artefacts occurring due to random movements is, however, more difficult to achieve through such methods

2.3.7.2 Magnetic susceptibility difference artefact

Magnetic susceptibility ( ) is the response of a substance to the applied magnetic field There are three levels of responses that have been described: diamagnetic, paramagnetic and ferromagnetic The diamagnetic response arises from the electrons surrounding the nuclei, while the paramagnetic response arises from molecules that have unpaired electrons Both these responses are relatively weak responses and

materials with such responses are safe to be used in MRI However, the ferromagnetic response is found in certain ferrous metals and the magnetic susceptibility due to this response is very large The relationship between magnetic susceptibility, external magnetic field and net magnetisation vector is expressed as the equation below:

2.3.7.3 Chemical shift

Chemical shift is the difference in precessional frequency conferred by the magnetic shielding effect of the electron clouds that surround protons within tissues, relative to that of a standard reference compound (in the case of protons tetramethylsilane ( TMS)) Basically, in MRI there are two sources of 1H nuclei, water and fat Water has two H atoms bonded to one oxygen atom, while fat has many H atoms bonded to

a long-chain carbon framework Due to this difference, protons from water have a different local magnetic field than protons from fat which is called „magnetic shielding‟ This magnetic shielding effect causes the protons from two sources to precess at different frequencies This, in turn, causes fat and water protons from the same tissue location map to different positions in the reconstructed image The difference of precessional frequency between water and fat at 1.5T is approximately 220Hz

MRI is designed to scan soft tissues utilizing 1H nuclei as the source of signal and, thus, is not routinely used for imaging of bones However, by using the signal generated from the surrounding soft tissue, bone outer geometry can be quantified from MRI images This will be discussed in more detail in Chapter 4

2.3.9 Advantages and current limitations of MRI

Absence of ionising radiation is an important advantage of MRI over CT, as this allows researchers to scan healthy human volunteers without exposing them to a high dose of ionising radiation However, MRI has some limitations compared to CT when used for scanning of long bones, such as longer scanning times, poor contrast

in certain anatomical regions, non-uniformity of the magnetic field, limited availability, and higher cost per scan

Longer scanning time is the most important limitation of MRI when it is used for scanning of clinical cases as well as for research As an example, in this study, scanning of a human lower limb with a modern 64 slice helical CT scanner takes less than ten seconds of scanning time, while an MRI scanner takes more than one hour for the same scan This longer scanning time of MRI makes the images of moving (breathing) body parts vulnerable to motion artefacts

2.3.9.2 Poor contrast in certain anatomical regions

The next important limitation of MRI is the poor contrast of MRI images in certain

anatomical regions of the bone (Figure 2.4) In the human body or other

mammalians (sheep), the diaphyseal region of long bones is covered mostly with muscles However, the distal and proximal regions of the bone, on the other hand, are mostly covered with ligaments, joint fluid, joint capsule and cartilage These different soft tissue types have different MRI properties and, depending on the chosen scanning parameters, some generate poor or no signal, thus making them indistinguishable from cortical bones (e.g ligaments, cartilage) Thus, the demarcation between such soft tissues and the cortical bone cannot be clearly defined and a complete 3D model is generated by making an educated guess or by interpolating the available data This educated guessing or interpolation of the regions introduces errors to the 3D models