Evaluation of chondral injuries using near infrared spectroscopy

uef.fi

PUBLICATIONS OF

THE UNIVERSITY OF EASTERN FINLAND Dissertations in Forestry and Natural Sciences

ISBN 978-952-61-2909-9 ISSN 1798-5668

Dissertations in Forestry and Natural Sciences

DISSERTATIONS | JAAKKO SARIN | EVALUATION OF CHONDRAL INJURIES USING NEAR INFRARED... | No 317

JAAKKO SARIN

EVALUATION OF CHONDRAL INJURIES USING NEAR INFRARED SPECTROSCOPY

THE UNIVERSITY OF EASTERN FINLAND

Currently, arthroscopic evaluation of chondral injuries is subjective and poorly reproducible. In this thesis, the potential of near infrared spectroscopy (NIRS), optical coherence tomography (OCT), and ultrasound

imaging was investigated for quantitative and reliable assessment of articular cartilage

and subchondral bone. The present findings support the clinical application of these techniques as they can address the limitations

of conventional arthroscopic evaluation enabling quantitative assessment of injury

severity and extent.

JAAKKO SARIN

PUBLICATIONS OF THE UNIVERSITY OF EASTERN FINLAND DISSERTATIONS IN FORESTRY AND NATURAL SCIENCES

N:o 317

Jaakko Sarin

EVALUATION OF CHONDRAL INJURIES USING NEAR INFRARED

SPECTROSCOPY

ACADEMIC DISSERTATION

To be presented by the permission of the Faculty of Science and Forestry for public examination in the Auditorium in Tietoteknia Building at the University of Eastern Finland, Kuopio, on October 12th, 2018, at 12:00 o’clock.

University of Eastern Finland Department of Applied Physics

Kuopio 2018

Grano Oy Jyväskylä, 2018

Editors: Pertti Pasanen, Matti Tedre, Jukka Tuomela, and Matti Vornanen

Distribution:

University of Eastern Finland Library / Sales of publications julkaisumyynti@uef.fi

http://www.uef.fi/kirjasto

ISBN: 978-952-61-2909-9 (print) ISSNL: 1798-5668

ISSN: 1798-5668 ISBN: 978-952-61-2910-5 (pdf)

ISSNL: 1798-5668 ISSN: 1798-5676

Author’s address: University of Eastern Finland Department of Applied Physics P.O.Box 1627

70211 Kuopio Finland

email: jaakko.sarin@uef.fi, jaakko.sarin@gmail.com Supervisors: Professor Juha Töyräs

University of Eastern Finland Department of Applied Physics P.O.Box 1627

70211 Kuopio Finland

email: juha.toyras@uef.fi Adjunct Professor Isaac Afara University of Eastern Finland Department of Applied Physics P.O.Box 1627

70211 Kuopio Finland

email: isaac.afara@uef.fi

Reviewers: Professor Sergio Fantini

Tufts University

Department of Biomedical Engineering MA 02155, Medford,

USA

email: Sergio.Fantini@tufts.edu Professor Gunter Spahn Jena University Hospital D-99817 Eisenach Germany

email: spahn@pk-eisenach.de

Opponent: Assistant Professor Gabriëlle Tuijthof Maastricht University

IDEE Engineering 6229 ER Maastricht Netherlands

email: gabrielle.tuijthof@maastrichtuniversity.nl

Jaakko Sarin

Evaluation of chondral injuries using near infrared spectroscopy Kuopio: University of Eastern Finland, 2018

Publications of the University of Eastern Finland Dissertations in Forestry and Natural Sciences, N:o 317

ABSTRACT

Arthroscopic evaluation of post-traumatic osteoarthritis (PTOA) is limited and sub- jective with the current diagnostic techniques visual evaluation and tissue palpation.

Reliable evaluation of cartilage defects and the surrounding tissue is crucial for selecting the optimal repair procedure. This requires introduction of novel quan- titative arthroscopic techniques, which could substantially enhance the currently limited diagnostics and, thus, improve patient care.

The aim of this thesis was to investigate the potential of various promising arthroscopic techniques for the assessment of chondral injuries. These techniques were near infrared spectroscopy (NIRS), optical coherence tomography (OCT), and ultrasound imaging. In particular, the potential of NIRS for evaluation of articular cartilage and subchondral bone integrity was thoroughly investigated. In studyI, two equine surgeons scored cartilage defects using a conventional arthroscope and by means of ultrasound and OCT imaging. Furthermore, cartilage stiffness measure- ments were performed with a hand-held arthroscopic indentation device (Artscan).

In addition, a custom algorithm was used to score the cartilage lesions visible in OCT images, and cartilage stiffness was evaluated with a standard table-top mate- rial testing device. In studyII, absorption spectra and biomechanical properties of equine articular cartilage were measured and multivariate regression modeling was employed to investigate their relationship. In studyIII, the combined potential of NIRS and OCT was investigated by measurements on equine samples, followed by quantitative microscopy to determine collagen orientation and content, and proteo- glycan (PG) content. In study IV, a novel NIRS probe was introduced for arthro- scopic measurements in equine jointsin vivoto evaluate cartilage and subchondral properties in a clinical setting. In studies III-IV, artificial neural networks (ANN) were utilized to relate spectra and reference properties.

In studyI, OCT supplemented with automatic scoring algorithm was the most reliable technique for scoring cartilage injuries; however, the advantage over the other techniques was not substantial. Overall, the findings suggest that imaging and quantitative analysis of the entire articular surface is critical to eliminate surgeon- related bias. In studiesII-IV, absorption spectra correlated significantly (p< 0.05) with cartilage biomechanical properties, PG and collagen contents, and collagen network orientation. Additionally, in studyIV, spectral information correlated with subchondral bone properties, and NIRS enabled simultaneous arthroscopic predic- tion of cartilage and subchondral bone properties. In study III, supplementing NIRS with automatic OCT-based characterization of cartilage lesions was found to enhance the prediction accuracy of the ANN models.

In conclusion, the findings in this thesis establish NIRS as a promising arthro- scopic technique for evaluation of cartilage integrity. However, supplementing NIRS with automatic OCT-based scoring of cartilage lesions further increases the potential of the technique.

National Library of Medicine Classifications: QT 34.5, WE 300, WE 348, WE 870, WN 180

Medical Subject Headings: Arthroscopy; Spectroscopy, Near-Infrared; Tomography, Op- tical Coherence; Ultrasonography; Microscopy; Cartilage, Articular/injuries; Osteoarthri- tis/diagnosis; Bone and Bones; Collagen; Proteoglycans; Biomechanical Phenomena; Elastic- ity; Multivariate Analysis; Neural Networks (Computer); Horses

Yleinen suomalainen asiasanasto: lähi-infrapunaspektroskopia; optinen koherenssitomo- grafia; ultraäänitutkimus; valomikroskopia; nivelrusto; luu; kollageenit; nivelrikko; biomekani- ikka; jäykkyys; joustavuus; monimuuttujamenetelmät; neuroverkot; hevonen

ACKNOWLEDGEMENTS

This study was carried out during the years 2014-2018 in the Department of Applied Physics at the University of Eastern Finland.

First and foremost, I would like express my sincere gratitude to my supervisors Professor Juha Töyräs, Ph.D., and Docent Isaac Afara, Ph.D., for giving me the op- portunity to join the research group, and for their professional guidance during this Ph.D. thesis. I am most grateful to my principal supervisor Juha for his enthusiasm and encouragement throughout my thesis and for always finding the time to tackle problems at hand. I also wish to thank my second supervisor Isaac for his support and guidance, especially in the early days of this project.

I wish to thank all the coauthors for their contributions in the publications with special thanks to Michael Amissah, Harold Brommer, Lassi Rieppo, and Nikae te Moller for their substantial contributions in the measurements. In addition, sin- cere gratitude goes for people in Utrecht, Netherlands, with whom I have had the privilege to work with in close collaboration.

I want to express my sincere thanks to the official reviewers of this thesis, Pro- fessors Gunter Spahn, Ph.D., and Sergio Fantini, Ph.D., for their professional review and constructive criticism.

Cheers go to my colleagues: Juuso, Tuomas, Mithu, and Jari, for all the guidance and distractions in the legendary office of ME358. I also wish to thank the coffee crew and rest of my colleagues in the research group of Biophysics of Bone and Cartilage (in alphabetical order): Aapo, Abi, Anni, Ari, Christina, Chuby, Elvis, Ervin, Gustavo, Hans, Janne, Jari, Jarkko, Jukka (×2), Kata, Kimmo, Lasse, Lauri, Markus, Mika, Mikael, Mikko (×3) Miitu, Mimmi, Moukku, Nina, Olli, Pete, Petro, Pia, Rami, Sami, Satu, Simo, and Weiwei.

I also wish to thank Academy of Finland, Kuopio university hospital (KUH), Orion foundation sr, and the finnish foundation for technology promotion for sup- porting my thesis and thus making it possible. In addition, thanks go to the Doc- toral programme in Science, Technology, and Computing (SCITECO) at University of Eastern Finland.

Thanks go to my parents, Hanna and Tarmo, and my brother Heikki for sup- porting me during my youth. Last and definitely not least, I want to present my utmost gratitude to Maija, who has endured my long working days like a champ, and brightened my days throughout this project.

In memory of my grandmother, Kuopio, October 12th, 2018

Jaakko Sarin

LIST OF PUBLICATIONS

This thesis consists of the present review of the author’s work in the field of op- tical spectroscopy of articular cartilage and the following selection of the author’s publications:

I J.K. Sarin, H. Brommer, D. Argüelles, P.H. Puhakka, S.I. Inkinen I.O. Afara, S. Saarakkala, and J. Töyräs, "Multimodality scoring chondral injuries in the equine fetlock jointex vivo,"Osteoarthritis and Cartilage25, 790-798 (2017).

II J.K. Sarin, M. Amissah, H. Brommer, D. Argüelles, J. Töyräs, and I.O. Afara,

"Near infrared spectroscopic mapping of functional properties of equine artic- ular cartilage,"Annals of Biomedical Engineering44, 3335–3345 (2016).

III J.K. Sarin, L. Rieppo, I.O. Afara, S. Saarakkala, and J. Töyräs, "Combination of optical coherence tomography and near infrared spectroscopy enhances deter- mination of articular cartilage composition and structure,"Scientific Reports7, 10586 (2017).

IV J.K. Sarin, N.C.R. te Moller, I.A.D. Mancini, H. Brommer, J. Visser, J. Malda, R.P. van Weeren, I.O. Afara, and J. Töyräs, "Arthroscopic near infrared spec- troscopy enables simultaneous quantitative evaluation of cartilage and sub- chondral bonein vivo,"Scientific Reports8, 13409 (2018).

Throughout the thesis, these papers will be referred to by Roman numerals.

AUTHOR’S CONTRIBUTION

The publications in this dissertation are original research papers on evaluation of cartilage and subchondral bone with near infrared spectroscopy (NIRS) and other quantitative arthroscopically applicable techniques,i.e., optical coherence tomogra- phy (OCT) and ultrasound imaging. In all the papers, the author participated in the study design and was the principal author.

In paper I, the author collaborated with veterinary surgeons H. Brommer and D.

Argüelles, on evaluation of cartilage defects, further developed the automatic scor- ing algorithm, performed the reference biomechanical measurements, and carried out all the data analysis.

In paperII, the author conducted NIRS and biomechanical measurements with M.

Amissah. Data analysis was conducted by the author.

In paperIII, the author conducted the OCT, NIRS, and digital densitometry mea- surements. L. Rieppo performed the polarized light and Fourier transform infrared microscopy measurements. The author conducted all the data analysis.

In paper IV, the author conducted the arthroscopic NIRS measurements with H.

Brommer and biomechanical measurements with N.C.R. te Moller. The author car- ried out the computed tomography measurements and all the data analysis.

TABLE OF CONTENTS

1 INTRODUCTION

2 KNEE JOINT

2.1 Articular cartilage... 3

2.2 Subchondral bone... 5

2.3 Post-traumatic osteoarthritis... 5

2.4 Current diagnostics and arthroscopic evaluation of cartilage injuries.. 6

3 INFRARED SPECTROSCOPY

3.2 Hardware... 12

3.3 Spectral preprocessing... 13

3.4 Analysis techniques and variable selection... 14

3.5 Near infrared spectroscopy of osteochondral tissues... 17

4 AIMS OF THIS THESIS

5 MATERIALS AND METHODS

5.2 Biomechanical testing... 23

5.3 Optical coherence tomography... 25

5.4 Ultrasound... 26

5.5 Arthroscopic evaluation... 27

5.6 Histology and quantitative microscopy... 27

5.7 X-ray microtomography... 28

5.8 Statistical analyses... 28

6 RESULTS

6.2 Near infrared spectroscopy of articular cartilage and underlying bone. 31 6.3 Combination of near infrared spectroscopy and optical coherence tomography... 33

6.4 Application of near infrared spectroscopy in vivo... 34

7 DISCUSSION

7.2 Near infrared spectroscopy on quantitative evaluation of tissue com- position and structure... 38

7.3 Effect of lesion severity on spectral response... 41

7.4 Arthroscopic near infrared spectroscopy... 41

8 SUMMARY AND CONCLUSIONS

BIBLIOGRAPHY

LIST OF ABBREVIATIONS

ANN Artificial neural network

BiPLS Backward interval partial least squares BMD Bone mineral density

BV Bone volume fraction

CARS Competitive adaptive reweighted sampling CCD Charge-coupled device

CI Confidence interval (95%) CV Coefficient of variation microCT X-ray microtomography DD Digital densitometry ECM Extra cellular matrix FCD Fixed charged density

FTIR Fourier transform infrared spectroscopy

GA Genetic algorithm

GAG Glycoaminoglycan

ICC Intra-class correlation coefficient ICRS International Cartilage Repair Society IVS Input variable selection

LD Lesion depth

LOO Leave-one-out

MC-UVE Monte Carlo uninformative variable elimination

MIR Mid infrared

MR Magnetic resonance

MSC Multiplicative scatter correction

NIR Near infrared

NIRS Near infrared spectroscopy

NRMSE Normalized root mean square error

OA Osteoarthritis

OCT Optical coherence tomography

OD Optical density

ORI OCT roughness index

PBS Phosphate-buffered saline PCA Principal component analysis

PG Proteoglycan

PLM Polarized light microscopy PLSR Partial least squares regression PTOA Post-traumatic osteoarthritis RMSE Root mean square error RMSEC RMSE of calibration RMSECV RMSE of cross-validation RMSEP RMSE of prediction ROI Region of interest

SD Standard deviation

SMI Structure model index SNR Signal-to-noise ratio SNV Standard normal variate

VCPA Variable combination population analysis

VIS Visible

LIST OF SYMBOLS

a Radius of indenter

A Absorbance

b0andb_k Regression constant and coefficients

c Speed of light

d Diameter

D Dark spectrum

EDyn Dynamic modulus

EEq Equilibrium modulus E_Inst Instantaneous modulus

Em Measured/Uncorrected modulus

F Indenter force

F_{Re f.plate} Reference plate force Gq Allowed energy level

h Planck’s constant

ht Tissue thickness

Iand I0 Light intensity after and before material Ii Identity matrix

J Jacobian matrix

l Traveled distance

k Force constant

n Number of measurement points

N Number of samples

p Level of statistical significance

q Quantum number

r Pearson correlation coefficient R Reflectance spectrum from standard R_C Indenter radius of curvature

S Sample spectrum

t Time

T Transmittance

vandvc Frequency and central frequency

w Weights

X Spectrum

X Average spectrum

yand ˆy Measured and predicted reference variable

e Strain

κ_method Inter-method reliability κ_intra Intra-observer reproducibility κ_inter Inter-observer reproducibility κandΩ Theoretical correction factors

λ Wavelength

µ Combination coefficient

µ_m Reduced mass of the diatomic molecule

ν Poisson’s ratio

ρ Spearman’s rank correlation coefficient σandσpre Stress and pre-stress

σa Activation function

χ Anharmonicity constant

1 INTRODUCTION

Articular cartilage is a specialized soft tissue covering the ends of articulating joints and together with synovial fluid enables near-frictionless motion of joints. Apart from providing this smooth motion, cartilage absorbs and distributes the mechan- ical loads experienced by the joint. Cartilage main constituents (i.e., water, proteo- glycans (PGs), and collagen) are essential for tissue function. Disruption of these constituents (e.g., as a result of trauma) can severely impair the function of carti- lage [1, 2].

Degenerative joint conditions, such as idiopatic osteoarthritis (OA) and post- traumatic OA (PTOA), are major causes of disability in human and other animals [3].

The initial stages of these forms of OA are characterized by changes in superficial cartilage: decrease in PG content, disruption of collagen network, and increased water content [4]. In impact loading of the joint, cartilage experiences excessive stress and strain, which may lead to localized chondral defects [4, 5]. These defects could jeopardize the integrity of the surrounding tissue to abnormally high stresses and strains, therefore leading to progressive degeneration (i.e., PTOA) [4].

The current diagnostics techniques, such as clinical examination, conventional radiography, and magnetic resonance imaging, are utilized in the initial diagnosis of joint conditions (e.g., ruptured ligaments or menisci). These conditions are treated in arthroscopic repair surgery, in which previously undetected chondral defects may be discovered as the current diagnostics techniques are only reliable in detecting late stages of the diseases [6–9]. In repair surgeries, an arthroscopic camera is utilized to visualize the joint cavity and a metallic hook is used to palpate tissue surface to eval- uate tissue condition. To evaluate the severity of cartilage lesions, several scoring systems, such as International Cartilage Repair Society (ICRS) scoring system [10], have been established [11, 12]. However, arthroscopic scoring of cartilage lesions is highly subjective [10, 13–15], which can lead to sub-optimal diagnosis and, thus, se- lection of non-optimal treatment. Therefore, adaptation of quantitative arthroscopic techniques could substantially enhance the evaluation of chondral defects.

Multiple techniques have been proposed for arthroscopic evaluation of cartilage defects, including optical coherence tomography (OCT) [16], ultrasound [17], and near infrared spectroscopy (NIRS) [6, 18, 19]. These techniques are complementary as OCT enables high resolution imaging of cartilage [20] and ultrasound allows imaging of subchondral bone [17], while NIRS provides information on tissue com- position [21, 22]. For thorough evaluation of cartilage defects, a combination of these techniques could be optimal. As scoring of cartilage lesion severity with high- resolution imaging modalities (i.e., ultrasound and OCT) is nearly as unreliable as with conventional arthroscope [23, 24], the process would benefit from automatic segmentation and scoring. In addition, in the initial stages of OA, compositional changes may occur prior to any visual cues. Therefore, sensitive evaluation of carti- lage composition could further enhance the detection of these initial stages and help to identify impaired cartilage,e.g., initial signs of PTOA. However, orthopedic sur- geons currently have no means for quantitative evaluation of cartilage composition without destructive approaches, such as biopsy extraction. Thus, introduction of a

quantitative arthroscopic technique, such as NIRS, could be highly beneficial. As a result, combination of OCT and NIRS could provide a comprehensive evaluation of cartilage and subchondral bone properties and could enhance the detection of initial signs of PTOA. This would have a high clinical significance,e.g., when conducting repair surgeries.

In addition to these arthroscopic techniques, several non-invasive techniques, in- cluding transillumination imaging, diffuse optical imaging, and fluorescence imag- ing, have been proposed for evaluation of arthritis. These techniques utilize visible and near infrared (NIR) spectral regions and have enabled, e.g., detection of joint inflammation and imaging with 1 mm resolution. However, the poor penetration depth of the light in biological tissues limits the exploitation of the techniques to small joints, such as fingers [25, 26].

As arthroscopic evaluation of cartilage defects is subjective and suffers from poor intra- and interobserver reliability [12–14,27–29], in studyI, the reliability of conven- tional arthroscopic scoring, and scoring reliability of ultrasound and OCT images were compared. In addition, chondral lesions visible in high-resolution OCT images were automatically scored using a custom algorithm (unbiased evaluator) to deter- mine the reproducibility and reliability of scoring by surgeons. Furthermore, the gold-standard histological scoring was performed by three experienced evaluators.

Severalin vitrostudies on feasibility of NIRS for evaluation of cartilage [6, 18, 30–

44] and engineered cartilage [21, 45, 46] have been published; additionally, a fewin vivostudies have been published [6, 19, 47–49]. These studies have introduced NIRS for evaluation of cartilage condition, but lack either in terms of clinical relevance or level of adapted analysis techniques. In this thesis, the potential of NIRS for clinical setting is investigated by adapting state-of-the-art multivariate models. In addition, the feasibility of the technique for assessment of the underlying bone properties was investigated, as only Afara et al has previously investigated the underlying bone properties in rat models of OA [32]. As NIRS is well-suited for point measurements, the feasibility of NIRS for mapping of equine cartilage functional properties is eval- uated in studyII. StudyIIIfocuses on investigating the combined potential of NIRS and OCT for comprehensive evaluation of cartilage properties, such as collagen con- tent and orientation. Previousin vivostudies have evaluated cartilage condition by using simplistic analysis techniques, such as ratio of two spectral peaks [6,19,47–49].

In this thesis, multivariate and shallow neural network approaches are adapted in studiesII-IVto model the relationships between spectra and reference properties of cartilage and subchondral bone. In studyIV, arthroscopic NIRS probe is introduced for simultaneousin vivoevaluation of cartilage and subchondral bone.

2 KNEE JOINT

The knee joint is the largest and most complex joint in the human body and consists of three bones (the ends of the femur and tibia, and patella), two menisci (medial and lateral), four ligaments, and two tendons. The ends of the long bones and patella are covered by articular cartilage, which ensures near-frictionless movement of the joint with the help of synovial fluid lubrication. Furthermore, the menisci, ligaments, and tendons assist in the stabilization of the knee and enable joint movement with the surrounding synovial membrane and muscles.

2.1 ARTICULAR CARTILAGE

Mammalian articular cartilage has remarkably similar biochemical composition over the species (e.g., mouse, human, equine, and elephant) [50] with equine and hu- man cartilage also alike in thickness [51]. Articular cartilage in the human knee is typically a 2 to 4 mm thick aneural and avascular tissue [50, 51] that relies on the diffusion from synovial fluid for uptake of nutrition [52]. Thus, articular cartilage has poor repair ability [52, 53], making the tissue susceptible to degeneration after trauma [3, 5, 54].

Composition and Structure

Articular cartilage consists mainly of extra-cellular matrix (ECM), which is primarily composed of collagen network and proteoglycans (PG) [1, 52], and interstitial fluid, which includes water and dissolved ions [1, 52]. In addition to these constituents, 1–5% of cartilage volume is occupied by chondrocytes [1, 2], which are cartilage cells responsible for synthesis of ECM, albeit with limited capacity to repair se- vere degeneration [1]. For the biomechanical function of the joint, the interaction between the ECM and interstitial fluid is essential [1, 52, 55]. Water is the main constituent of cartilage, contributing up to 80% of wet weight, whereas structural macromolecules (i.e., collagens, PGs, and non-collagenous proteins) contribute 20 to 40% of the wet weight [1, 2]. For articular cartilage dry weight, the collagens, PGs, and non-collagenous proteins contribute 60%, 25–35%, and 15–20%, respectively [1].

Several types of collagen exists in articular cartilage, including types II, VI, IX, X, and XI, with type II being the most abundant (up to 95%) [1,2]. Collagen is structured as long fibrils extending throughout the cartilage from cartilage surface to the calcified cartilage [52]. These collagen fibrils form the framework of articular cartilage within the highly negatively charged PGs are trapped [52]. PGs consist of a protein core and a single or multiple glycoaminoglycan (GAG) chains (e.g., hyaluronic acid and chondroitin sulphate) [1, 2], which are responsible for fixed charge density (FCD) of cartilage [56]. The FCD plays a vital role in biomechanical function of cartilage [56].

Articular cartilage is conventionally divided into four layers based on the struc- tural organization of the collagen fibers from cartilage surface to cartilage-subchond- ral bone interface: superficial zone, middle zone, deep zone, and calcified cartilage (Figure 2.1) [1, 52, 57]. The superficial zone is the thinnest layer (5–20% of cartilage

thickness [1, 52]), in which the collagen fibrils are aligned parallel with cartilage surface. The dense collagen network in the superficial zone contributes more to tis- sue tensile stiffness and strength compared to collagen fibrils in the other zones [1].

In the middle zone, the collagen fibrils are randomly oriented [1, 52]. This zone has relatively lower concentration of water and collagen compared to the superficial zone, whereas the PG concentration is higher [1, 52]. In the deep zone (up to 85% of cartilage thickness), collagen fibrils are thicker than in the other zones and are ori- ented perpendicularly relative to the cartilage surface [52, 58]. Towards the calcified cartilage, PG concentration further increases, whereas water content decreases [1].

The region differentiating the deep zone and calcified cartilage is referred to as the tidemark [1, 52]. The calcified cartilage plays an important role of securing the car- tilage matrix to the underlying bone by anchoring the collagen fibrils of cartilage.

The shape and distribution of chondrocytes varies as a function of depth. In the su- perficial region, the cells are shaped as ellipsoids oriented along the collagen fibrils.

In the middle zone and deep zone, the chondrocytes are shaped as spheroidals and they form long columns when approaching the calcified layer along the orientation of collagen fibril network [1].

Figure 2.1:Basic anatomy of human knee joint along with structure and composi- tion of articular cartilage and underlying bone.

Function

The main function of articular cartilage is to distribute stresses and strains [1, 2, 52].

In static joint loading (e.g., during standing), cartilage continues to deform until equilibrium between the deforming force and internal resisting forces of cartilage is reached [2]. During the loading, interstitial fluid pressure increases, causing fluid to slowly flow out from cartilage [52]. Upon removal of the load, the osmotic pressure due to the increased FCD causes the fluid to flow back into the tissue; furthermore, the collagen network restricts cartilage from swelling beyond the original form. In high rate dynamic loading, the low permeability of cartilage resists deformation, making cartilage almost incompressible, with high stiffness [1, 52]. Overall, the col- lagen network has a substantial effect on the dynamic properties of cartilage, while PGs contribute to time-dependent deformation of the tissue [59].

2.2 SUBCHONDRAL BONE

Subchondral bone plate and subchondral trabecular bone are the underlying calci- fied tissues that provide the base for articular cartilage. These tissues differ in their morphology and mechanical properties and have been shown to respond differently in OA progression [60].

Structure, Composition, and Function

Subchondral bone plate has a compact structure, higher mineral content, and lower porosity compared to the subchondral trabecular bone [61, 62]. These differences in bone distribution and alignment substantially contribute to their mechanical prop- erties; although, they possess similar composition and material properties [63]. In epiphysis of bones, subchondral bone plate and subchondral trabecular bone help to absorb the impact loads of joints [63, 64] due to their ability to deform under loads. [63]

Bones consist of organic material (25% of wet weight), inorganic material (65%), and water (10%) [63]. Collagen is the primary (90%) organic material (95% of type I with small amounts of types V and XII), providing bone with its tensile strength, structural integrity, and ductility [63, 65, 66]. The remaining 10% of the organic material is formed of non-collagenous proteins, which are responsible for organiz- ing collagenous matrix, and regulating growth rate and stability of mineral crys- tals [63, 67]. The inorganic material (the mineral phase) serves as ion reserve (e.g.

calcium, phosphorus, sodium, and magnesium) and provides bone most of its stiff- ness and strength [63]. Calcium-phosphate crystals (i.e., hydroxyapatite) along with the organic matrix of bone are essential for bone to cope with the forces imposed by joint loading [63]. In bone tissue, water is mostly found within pores and as bound to the matrix. The pore water (i.e., free or mobile water) can freely move according to the pressure gradient developed by joint loading and is mainly found in Haver- sian and Volkmann’s canals [68]. The bound water is attracted by the hydrophilic residues of collagen molecules, which bound the water loosely or tightly [68]. The bound water has an essential role in bone mineralization, which has a substantial effect on bone mechanical properties [68, 69]. In contrast to articular cartilage, bone has an effective repair ability due to its active blood supply [63].

2.3 POST-TRAUMATIC OSTEOARTHRITIS

Osteoarthritis (OA) is a prevalent and disabling disease causing substantial so- cioeconomic burden [70]. While OA is considered as a late-stage condition for which disease-modifying opportunities are limited (e.g., joint replacement), the dis- ease conventionally develops over decades, thus offering a long period to alter its progression [71]. Unlike idiopathic OA, which is age-related, post-traumatic OA (PTOA) is a form of OA which is initiated by a sudden trauma,e.g., intra-articular fracture, joint dislocation, torn meniscus, or anterior cruciate ligament tear. Al- though limited statistics are available on the prevalence of PTOA, roughly 5.6 mil- lion cases of PTOA are reported annually in the United States alone, with 12.5% of these being knee related injuries [72]. These traumas often lead to rapid joint degen- eration in a high proportion of patients [73,74]. In these cases, local cartilage defects will subject the surrounding cartilage to excessive stresses and strains, thus further damaging the surrounding tissue. While OA mainly affects the elderly population,

PTOA affects people of all ages, thus offering a unique opportunity to substantially reduce the number of joint replacement cases via early diagnosis. The initial signs of PTOA are similar to those of OA, including disruption of superficial collagen network and PG loss [5]. These alterations are followed by increased water content, resulting in cartilage softening [5]. The initial structural and compositional changes in cartilage may occur prior to any visible changes, thus making visual inspection using a conventional arthroscopy unreliable.

2.4 CURRENT DIAGNOSTICS AND ARTHROSCOPIC EVALUATION OF CARTILAGE INJURIES

Conventionally, initial diagnosis of OA-related diseases is performed by the clinician via physical examination and based on symptoms described by the patient. How- ever, joint injuries or degeneration may also be asymptomatic, which delays the diagnosis [75]. The most common symptoms of late stage OA are joint stiffness and pain [5]. The diagnosis is conventionally confirmed via plain radiographs [76–78]

with the radiographs being generally scored according to Kellgren-Lawrence grad- ing system [79]. Due to the poor soft tissue contrast of conventional radiography, the grading is based on joint space narrowing and bone changes. These conditions are, however, associated with late stages of OA. Therefore, plain radiographs are insufficient for diagnosing initial signs of OA [78]. The contrast between the joint space and cartilage can be enhanced in X-ray tomography by injecting contrast agent in the synovial cavity (contrast-enhanced computed tomography) [80], thereby visu- alizing the cartilage surfaces of the joint and enabling detection of cartilage loss. In addition, magnetic resonance (MR) imaging may be conducted as it enables evalua- tion of soft tissues (e.g., cartilage and meniscus). However, MR imaging suffers from relatively poor resolution and has limited availability due to relatively long imag- ing times, thus limiting the feasibility of the technique in diagnosing early stage degeneration [81, 82].

Ligament and meniscal tears are conventionally treated in arthroscopic repair surgery. In these surgeries, previously undiagnosed cartilage lesions may be discov- ered, which may require surgical repair to prevent the initiation or progression of PTOA. Current arthroscopic evaluation of cartilage lesions relies on visual inspec- tion with an arthroscopic camera and tissue palpation using a metallic hook. While subjective, the evaluation of cartilage softening presents the earliest detectable clini- cal sign of pre-OA changes, also known as chondromalacia or chondrosis [11,71]. To evaluate visible cartilage lesions, several arthroscopic scoring systems have been in- troduced, such as the Internal Cartilage Repair Society (ICRS) and Outerbridge scor- ing systems, which are applied in quantifying the severity of cartilage lesions [10,11].

In ICRS scoring system, the severity of cartilage lesion is described with ICRS0 as normal cartilage, ICRS1 as superficial softening, fibrillation, and fissures, ICRS2 as defects extending less than 50% of cartilage thickness, ICRS3 as defects extending deeper than 50% of cartilage thickness but not extending into subchondral bone, and ICRS4 as defects extending into subchondral bone [10]. However, quantify- ing the aforementioned differences with current arthroscopic tools is poorly repro- ducible [13–15], requiring adaptation of novel diagnostic techniques.

Several quantitative techniques, such as ultrasound and optical coherence to- mography (OCT), have been proposed for accurate and objective assessment of car- tilage integrity. Ultrasound has been applied for cartilage imaging non-invasively

via linear transducers [83]. In addition, intravascular ultrasound catheters have been utilized in arthroscopies [17, 84]. Similarly, OCT has been applied arthroscopically by utilizing intravascular catheters [23, 85], which enable imaging in narrow joint cavities. These techniques provide superior resolution compared to conventional imaging modalities (i.e., computed tomography and MR imaging) and evaluation using an arthroscope. In addition, ultrasound and OCT techniques have been in- troduced for quantitative analysis of cartilage properties,e.g., to evaluate cartilage biomechanical properties [17, 86]. To determine cartilage true mechanical properties in arthroscopyin vivo, arthroscopic indentation systems (e.g., Artscan [87, 88] and ACTAEON Probe [89]) have been introduced. However, the measurements with these systems were user-dependent with relatively poor reproducibility. In addition, an ion-streaming potential probe (arthro-BST) has enabled evaluation of cartilage stiffness and healthex vivo [90, 91]. Near infrared spectroscopy (NIRS) was intro- duced in the last decade for quantitative evaluation of cartilage in arthroscopy [47].

The technique enables determination of cartilage composition and could thus sub- stantially enhance the outcome of conventional arthroscopic evaluation of joint in- tegrity. However, the full-potential of the technique for in vivo assessment of joint integrity has not been explored, including adaptation of novel analysis techniques (e.g., partial least squares regression (PLSR), artificial neural networks (ANN), and sophisticated variable selection techniques), as well as evaluation of the health of tissue surrounding cartilage lesions.

3 INFRARED SPECTROSCOPY

Infrared spectroscopy is based on the vibrational and rotational transitions of atoms and molecules in temperatures higher than absolute zero. These vibrations are bond-specific and can, thus, enable identification of molecules [92]. Mid infrared (MIR) spectroscopy has been applied for evaluation of tissues due to the strong fundamental vibrations within this spectral region [93, 94]. However, in the visible (VIS) and near infrared (NIR) spectral regions, only the weaker overtone and com- bination vibrations can be observed [95, 96]. This results into overlap of spectral features, which substantially limits the use of these regions [96]. Only recently has NIR spectroscopy been utilized due to advances in chemometrics,e.g., in agriculture to determine wheat quality [97, 98] and in biomedical engineering to evaluate tissue properties [32, 41, 99].

3.1 THEORETICAL BACKGROUND

Light can be described using the wave-particle duality concept, in which the behav- ior of a photon is not only explained as a particle but also as a wave. The energy of a photon is defined by:

E=hv= ^hc

λ, (3.1)

whereh is Planck’s constant,vis frequency,cis velocity, andλis wavelength. The energy of a photon is, therefore, dependent on the wavelength. Based on photon energy and the magnitude of its contribution, VIS (λ = 0.4–0.75 µm), NIR (λ = 0.75–2.5 µm), and MIR (λ = 2.5–1000 µm) spectral regions have been introduced.

These regions include bond-specific vibration frequencies, commonly presented as wavenumbers (v, cm⁻¹). In these spectral regions, absorptions arise due to bending and stretching bond vibrations in molecules [92, 96]. These vibration frequencies describe the potential energy difference between two energy levels and each of these levels is described by a quantum numberq= 0, 1, 2, ... (Figure 3.1).

The transitions between the ground level (i.e., level 0) and level 1 are called the fundamental transitions (appear in MIR region), whereas transitions from 0 to q= 2, 3, 4, ... are called overtones and other transitions are called hot transitions (ap- pear in VIS and NIR region). The vibration frequency of overtones can be roughly estimated as the product of the fundamental transition frequency (in wavenumbers) and an integer; however, the overtone frequencies can be determined more accu- rately. Based on the harmonic diatomic oscillator model (Figure 3.1), the allowed energy level (expressed in cm⁻¹) can be derived as

Gq = (_q+¹

2)_v0= (_q+¹ 2) ¹

2πc s k

µ_m, (3.2)

whereqis the vibrational quantum number (0, 1, 2, ...),kis the force constant of the bond, andµm is the reduced mass of the diatomic molecule [96]. Furthermore, the

harmonic oscillator model is only valid for neighboring energy levels (∆q=±1) and does not account for repulsive forces between the atoms or dissociation of strongly extended vibration bonds [92, 96]. Morse potential (Figure 3.1), however, accounts for the anharmonicity and bond breaking [92, 96]. The allowed energy levels are described by:

Gq = (q+¹

2)v0−(q+¹

2)^2χv0, (3.3)

where χis the anharmonicity constant [96, 100]. The frequencies of overtones (vq) can be determined by [100]

vq =Gq−G0

=qv0(1−^χ(q+1))

= ^qv1(1−^χ(q+1)) 1−2χ ,

(3.4)

where q = 2, 3, 4, ... andv₁ is the fundamental vibration frequency (transition: 0

→ 1). Additionally, the effects of rigid rotor [92], centrifugal distortion [92], and rotational-vibrational coupling [101] can be determined with the molecular-specific constants of rotation, centrifugal distortion, and vibration-rotation interaction, re- spectively, along with the rotational quantum number.

Figure 3.1:Schematic of harmonic oscillator potential and morse potential.

In addition to overtones (λ= 0.7–1.8 µm), combination bands (λ= 1.3–2.7 µm) can be observed in the NIR spectral region (Table 3.1) [96]. The association of over- tones and these combinations bands results in overlapping spectral features and, thus, substantially decreases the specificity of absorption in this spectral region. The relative contribution of absorption bands is strongest in the MIR region (fundamen- tal vibrations). Furthermore, with increasing overtones (decreasing wavelength), the relative absorbance is weaker(∼10–100 times weaker) [95]. Recent advances in computer hardware, regression techniques, and availability of computational power have accelerated the adaptation of these spectral regions in multiple applications.

The optical response of a sample can be expressed as absorbance, transmittance, or reflectance spectroscopy, of which reflectance spectroscopy was utilized in this

thesis. In reflectance spectroscopy, light is transmitted into a sample, and the re- flected and back-scattered light is collected (Figure 3.2). Rest of the light is either absorbed by the sample or transmitted through the sample. To determine spectral properties of a sample, Beer-Lambert’s law, also know as Beer’s law [102], can be applied. The law describes the decrease in light intensity as it traverses through the tissue and is described by:

I= I010^−µl, (3.5)

where I is light intensity after material, I0 is the initial light intensity, µ is the material-specific absorption coefficient, and l is the distance traveled through the material. Transmittance (T) or absorbance (A) of the material is determined with

A=−log₁₀T=−log₁₀( ^I

I₀). (3.6)

Table 3.1:Common NIR bands [95, 96]

Wavelength (µm) Other

0.78–0.85 Third overtone N–H stretching 0.85–0.95 Third overtone C–H stretching

0.95–1.10 Second overtones of N–H and O–H stretching 1.10–1.23 Second overtone C–H stretching

1.30–1.42 Combination C–H stretching

1.40–1.55 First overtones of N–H and O–H stretching 1.65–1.80 First overtone C–H stretching

1.90–2.00 Second overtones of O–H bending and C=O stretching 2.00–2.20 Combination N–H stretching, combination O–H stretching,

Second overtone N–H bending

In reflectance spectroscopy, the absorbance is based on the reflection and scatter- ing in the sample, and can be determined with

A=−log₁₀(^S−D

R−D), (3.7)

whereS is the scattered and reflected light collected from a sample,Dis the spec- trum from a dark reference standard, and R is the spectrum from a reflectance standard. The dark reference is determined to account for hardware-related noise, whereas the reflectance standard is an optimal specular reflector that conventionally reflects over 99% of light across the specified wavelength region.

3.2 HARDWARE

A conventional reflectance system consists of two components: a light source and a spectrometer. Additionally, fiber optic probes can be utilized to effectively propagate the light into a sample and to collect the reflected and scattered light, therefore substantially enhancing signal-to-noise ratio (SNR) of data acquisition. These probes can be designed for different applications, such as extremely durable probes for industrial applications or sterilizable probes for medical use.

Optical spectrometers are commonly referred to as spectrometers; although, sev- eral other types of spectrometers (i.e., mass spectrometers, time-of-flight spectrom- eters, and magnetic spectrometers) are available. These aforementioned spectrom- eters are commonly applied in a laboratory environment and are not suitable for in vivo analysis. In the context of this thesis, a spectrometer is referred to as an instrument for measuring light intensity across a wide wavelength region. In these spectrometers (Figure 3.2), light is guided through a narrow slit, which is usually followed by a collimating mirror, a fine grating, and a detector or a detector array.

To achieve a higher resolution, the slit can be narrowed and the lines of the grat- ing increased. However, reducing the slit size will result in signal loss which is usually compensated by longer exposure times. Forin vivoapplications, spectrom- eter systems should be optimized to enable measurements with short duration and sufficient SNR.

As a light source, tungsten-halogen lamps have been broadly applied in re- flectance measurements. These sources emit light in the VIS and NIR spectral re- gions, with no emission in the harmful ultra-violet region as opposed to deuterium light sources.

Figure 3.2: Schematic representation of conventional reflectrance spectroscopic measurement along with basic components of a spectrometer. Optical fibers trans- mit light into a sample from which the scattered and reflected light is collected and transmitted to a spectrometer. In the spectrometer, light passes through a slit, fol- lowed by reflections from a collimating mirror, a reflection grating, and a focusing mirror. Finally, the signal is measured at the detector and displayed.

3.3 SPECTRAL PREPROCESSING

Spectral preprocessing is essential as signals often include noise, such as baseline fluctuation due to temperature differences and detector nonlinearity in different spectral regions. In reflectance spectroscopy, the environment should be controlled and any stray light must be eliminated. Additionally, fluctuations in wavelength- specific response of sensors and output intensity of light source will effectively result in different SNR across the spectral regions. In particular, spectral measurement at the fringes of the wavelength range of the detector usually have poor SNR due to geometry of the spectrometers (Figure 3.3A). Therefore, using data from these spectral regions may impair the reliability of predictions in multivariate modeling as the models interpret systematic variations in the sample spectrum as a consequence of variation in the sample.

Several methods have been used for spectral preprocessing, such as scatter cor- rections and spectral derivates [103]. The principle of scatter correction algorithms, e.g., multiplicate scatter correction (MSC) and standard normal variate (SNV), is to correct data artifacts and imperfections [103]. In MSC [104], constant termsb₀ and b₁are determined for all wavelengths and each sample by

X_i=b₀+b₁X+e, (3.8)

whereX_iis a sample spectrum and Xthe average spectrum, and thus the corrected spectrum is calculated as:

Xi,MSC = ^Xi−b0

b₁ . (3.9)

With SNV [105], the corrected spectrum is calculated using:

X_i,SNV= ^Xi−X_i

SD(Xi)^{, (3.10)}

where SD is the standard deviation of the spectrum.

Derivative algorithms are generally applied to enhance spectral features [103, 106]. The first derivative preprocessing removes the baseline offset, whereas the second derivative removes the dominant linear term (Figure 3.3) [106, 107]. For reli- ability of calibration models, sufficient preprocessing is essential as too little smooth- ing may not effectively eliminate the background or hardware-related noise, while too much preprocessing may result in loss of essential spectral information. In the widely applied Savitzky-Golay derivative preprocessing algorithm [108], a polyno- mial function is fitted to the raw spectrum with specific window width, followed by analytically determining the derivatives [103, 108]. This operation is then sequen- tially performed for all window locations.

Figure 3.3: Two raw spectra (grey) and two examples (black line and dotted blue) of smoothed raw spectra using the Savitzky-Golay filter (A), first derivative spectra (B), and second derivative spectra (C). In the raw spectra (A, grey), substantial signal variation is observable in the outermost regions of the spectrum.

3.4 ANALYSIS TECHNIQUES AND VARIABLE SELECTION

Both univariate and multivariate approaches have been applied in spectroscopy [95].

However, with the lack of specificity and heavily overlapping overtones in the NIR region, univariate analysis often results in unreliable models; nevertheless, it has been applied for cartilage evaluation by Spahnet al[19,47–49]. Utilization of the NIR spectral region has been vastly under-developed by the scientific community until recently, as it has now gained considerable attention in multiple fields requiring fast analytical solutions. This is due to the introduction and application of multivariate analysis, enabling swift evaluation of complex non-linear data. Multivariate model- ing is conventionally applied in classification or prediction problems to understand relationships between non-linear multivariate data and reference parameters. For multivariate modeling, a large number of observations is required in order to train and develop robust and well-generalized models.

Partial least squares regression

Partial least squares regression (PLSR) and principal component regression (PCR) are the most commonly applied chemometric techniques in NIR spectroscopic anal- ysis [96, 109]. Conventionally, PLSR is more robust when compared to PCR due to two crucial differences. Firstly, in a process of deflation, each PLS factor is subtracted from the predictor variables (spectral data), therefore enabling development of ro- bust models by maximizing the covariance structure in the data. Secondly, PLSR takes into account the variation of the dependent or response variable (y, reference).

The method was first introduced by Woldet al[109] to describe complicated multi-

variate systems by a sequence of simple least squares regressions [96]; furthermore, it can efficiently process highly multi-collinear variables. In multivariate analysis, the goal is to predict a reference variable (y) from a set of complex multivariate predictor variables (X) using:

ˆ

y_i =b₀+

∑

X_ikb_k, (3.11)

whereX_ikis absorption at wavelength (k) andb₀is a regression constant andb_k re- gression coefficients that are estimated statistically based on calibration data: spectra (X) and reference variables (y).

In PLSR modeling, choosing the optimal number of PLS factors is essential. If too many factors are included in the model, this may result in overfitting, in which the model is tailored to the training data and does not generalize well with new samples.

Furthermore, with too few PLS factors, key variance and important relationships between predictor and response variables may be unaccounted for in the model, therefore limiting the predictive capability of the model.

Validation of the accuracy and performance of PLSR models generally involves one of two techniques, namely cross-validation or (independent) test set validation.

Cross-validation techniques, such as leave-one-out (LOO) ork-fold, enable the de- termination of the optimal number of PLS factors and, thus, ensure generalization of the model. In LOO cross-validation, a single sample is iteratively excluded from modeling and a value is predicted with the model based on other samples. Ink-fold cross-validation, a similar exclusion is performed by dividing the data intokgroups.

With independent test set validation, data is split in two groups: calibration and validation groups (e.g., 66% and 33%). A model is built with the calibration group with internal cross-validation (LOO or k-fold), which is followed by validation of the model performance using the independent test set. The prediction performance of models is determined with the minimum root mean square error (RMSE).

RMSE= s1

n

∑

(y_i−yˆ_i)², (3.12) wherey_iis measured reference value and ˆy_ipredicted value. The RMSE parameter has several variations, which depend on the model and cross-validation used. The three most common variations are RMSE of calibration (RMSEC) describing the error of the calibration model, the RMSE of cross-validation (RMSECV) describing the error of iteratively excluded samples, and the RMSE of prediction (RMSEP) the error of an independent test set. Performance of models can be also described with other metrics, such as Pearson correlation coefficient (r).

Artificial neural network

Artificial neural networks (ANN) were originally designed to mimic how the human brain processes information. Due to the ability of neural networks to model non- linear relationships and the breakthrough in efficiently implementing them, they have been applied in several fields for solving complex problems, with high accu- racy. Modeling with ANN can be unsupervised or supervised, where the latter

involves information on the reference values. ANNs are conventionally categorized into shallow or deep networks, including a single or multiple hidden layers, respec- tively. Each hidden layer includes artificial neurons which receive a set of weighted inputs, process the sum and apply an activation function (σa), and then pass the result onwards to the next layer. The most common activation functions are

linear:σa(x) =x sigmoid:σ_a(x) = ¹

1−exp(−x) hyperbolic tangent:σ_a(x) =tanh(x).

(3.13)

The objective in ANN is to find a set of weights (w) that minimizes the sum of squared error (E(X,w)) of predictionei =yˆi−yi

E(X,w) =

∑

i

e²_i =

∑

i

(σ_a(w^TX)−yi)². (3.14) Several algorithms are available for optimizing the weights, including steep- est descent algorithm, Newton’s method, Gauss-Newton’s algorithm, and Leven- berg-Marquardt algorithm [110, 111]. The Levenberg-Marquardt back-propagation algorithm incorporates the steepest descent method and the Gauss-Newton algo- rithm [112], and is generally favored due to its stable and fast optimization. The weights are updated as follows

w_k+1=w_k−(J_k^TJ_k+µIi)⁻¹J_ke_k, (3.15) where J is the Jacobian matrix, µ is the always positive combination coefficient, and Ii the identity matrix. When the combination coefficient is very small (µ ≈0), the algorithm behaves similarly to Gauss-Newton algorithm, whereas with very big combination coefficient, the steepest descent method is used [112]. In the modeling, deep neural networks are considered to be better in generalization of the problem but require substantial amount of data and computational resources. In addition, adapting unnecessarily complicated networks may result in overfitting and, thus, reduce model reliability when introduced with new data. During initialization of model building, the originally introduced weights should be randomized and small (close to zero) to break symmetry and avoid saturation, respectively [113]. Addition- ally, initial weights have a substantial effect on the convergence speed of learning.

Variable selection

Multivariate data (e.g., NIR spectra) are often complex and include dispensable pa- rameters; thus, only the variables that enhance model reliability should be retained.

Variable selection techniques can be divided into two categories: techniques that de- termine the most relevant variables and techniques that eliminate the least relevant non-contributing variables. Conventionally, the technique based on selection of the most relevant variables achieves better results swiftly due to the lower amount of variables processed. Currently, several variable selection techniques are available, such as Monte Carlo uninformative variable elimination (MC-UVE), competitive adaptive reweighted sampling (CARS), variable combination population analysis (VCPA), backward interval partial least squares (BiPLS), genetic algorithm (GA),

and jack-knife [114]. In addition, exhaustive variable selection techniques (e.g., for- ward selection) can be applied. In forward variable selection technique, univariate models are built upon which the single most reliable variable is selected based on model performance. The selection is continued by iteratively identifying the next most reliable variable by building a bivariate model with retainment of the first variable [115]. However, exhaustive search that evaluates all variable combinations is the only approach that guarantees the optimal model performance [116]; there- fore, this is extremely time consuming.

3.5 NEAR INFRARED SPECTROSCOPY OF OSTEOCHONDRAL TIS- SUES

Arthroscopic repair surgery is commonly performed to treat joint injuries. However, its potential for successful repair is substantially limited by the weakness of current arthroscopic diagnostic techniques. These include qualitative visual evaluation of joint via arthroscope and tissue palpation with a metallic hook. Only in the last decade has NIRS been introduced with promising results for quantitative evalua- tion of knee tissues: articular cartilage [6, 18, 19, 30, 31, 33–35, 37, 40, 41, 43, 44, 47–49], subchondral bone [32], and meniscus [99, 117]. NIRS has been applied in mul- tiple ex vivo studies using human [39, 41] and animal samples, including bovine [33–35, 39, 40, 42, 43, 118], rat [31, 32, 37, 44], and sheep [30], thus providing valuable information for the clinical application of the technique. Bovine cartilage is widely used in testing the capability of NIRS for evaluation of cartilage integrity; this is because bovine cartilage thickness is similar to human’s. In clinical sense, equine would be a more appropriate animal model due to financial implications of cartilage lesions, for example, in racing horses.

Near infrared spectra has been shown to correlate with cartilage thickness [31, 33, 41], biomechanical properties [18, 30, 34, 41, 43], water content [30, 35], collagen content [45], and PG content [40, 41]. In addition to these functional and compo- sitional properties, NIR spectra has been correlated with severity of cartilage de- fects determined via conventional scoring systems including Mankin histological score [30, 31, 37, 44], ICRS score [18, 19, 47–49], and KOOS-score [6]. Additionally, investigations on wavelength-specific light propagation [42] and contribution of un- derlying bone [39] from NIR reflectance spectra have been conducted. However, no study has investigated the feasibility of NIRS for characterization of different cartilage layers or collagen orientation.

The research on OA-related diseases, such as PTOA, has mainly focused on the role of articular cartilage; however, several studies have suggested the underlying bone to have a substantial role on the progression of the disease [119,120]. Therefore, accurate and reliable evaluation of the underlying tissue (i.e., subchondral bone plate and subchondral trabecular bone) would be of high clinical significance. Afara et al [32] showed that the NIR spectral region (4000 - 12500 cm⁻¹) correlates with subchondral bone mineral density (BMD) and bone volume fraction (BV) in a rat model; however, the thickness of rat cartilage is substantially thinner than that of equine or human cartilage [50]. In addition, McGoverinet al[39] showed that NIR spectra measured from the direction of cartilage surface also includes contributions from the underlying bone, thus suggesting that the technique could be feasible for assessing bone integrity through relatively thick cartilage.

In addition to the NIR spectral region, VIS and MIR have been applied for car-

tilage evaluation [121, 122]. With respect to penetration of light into biological tis- sues [35], light in the MIR region is restricted to the superficial layer of cartilage and is, thus, mainly applied in laboratory environment for histological imaging.

Several studies have adopted VIS region for evaluation of tissues [41, 121] as the VIS light penetrates deeper into soft tissues. Thus, spectral measurements in this region would include contributions from multiple tissues. Furthermore, to separate tissue contributions, a detailed understanding of light penetration into soft tissues is required.

Interestingly, most laboratory studies have utilized PCR or PLSR to investigate the relationships between NIR spectra and the reference parameters. The analy- ses have mainly focused on the comparison of spectral regions or spectral shapes, with very few studies utilizing sophisticated variable selection techniques [114,123].

Additionally, shallow neural networks could provide substantial benefit over these conventional multivariate techniques [124, 125].

Arthroscopic NIRS was firstly introduced by Spahn et al[47] for evaluation of OA by prediction of ICRS scores. Otherin vivo studies by Spahn et al [19, 48, 49], Hofmannet al[6], and Martickeet al[18] have shown arthroscopic predicted ICRS scores to have good reliability when compared to ICRS scores based on MRI and X-ray images. However, these conventional imaging modalities are suboptimal for scoring cartilage defects. Furthermore, these studies are based on ineffective and simplistic determination of the ratio between two spectral peaks and focus on eval- uation of cartilage lesion severity. More recent ex vivostudies have adopted novel multivariate approaches to determine cartilage properties via NIRS [44]. These more sophisticated approaches would enable the evaluation of the cartilage surrounding the defect and, therefore, to evaluate the extent of the post-traumatic degeneration around the original lesion.

As no cure currently exists for OA or PTOA, several cartilage repair and regen- eration techniques are being developed and optimized. However, current arthro- scopic diagnostics measures are subjective and unreliable and thus not feasible for delineation of injured area to be repaired or monitoring tissue healing after repair surgery. Therefore, several studies have investigated the potential of NIRS to moni- tor engineered cartilage with positive outcomes [21, 38, 45, 46].

Overall, NIRS has shown promise for evaluation of cartilage based onex vivo findings in animal models. However, adaptation of recently introduced advanced analysis algorithms forin vivo environment should be investigated to demonstrate the clinical potential of the technique. Additionally, as NIRS provides quantitative information on cartilage composition, the technique would benefit from combina- tion with a high resolution imaging modality, such as OCT, for a holistic optical diagnosis of cartilage integrity during arthroscopic surgery.

4 AIMS OF THIS THESIS

Cartilage defects, ruptured menisci or ligaments may be treated via arthroscopic intervention. The current highly-subjective arthroscopic evaluation of the severity and extent of joint or cartilage injury is performed based on visual evaluation and by palpating tissue surface with a metallic hook. The diagnostics would benefit from introduction of novel quantitative arthroscopic techniques which could enable, for the first time, the surgeons to objectively evaluate the extent of compromised tissue and, thus, improve patient care. Therefore, the aims of this thesis were:

• To investigate the reproducibility of scoring cartilage lesions based on con- ventional arthroscopic visualization, ultrasound, and OCT, and develop an automated algorithm to score cartilage lesion severity based on OCT imaging

• To investigate the relationship between NIR absorption spectra and functional properties of articular cartilage

• To investigate the combined potential of OCT and NIRS in the evaluation of cartilage composition and structure

• To design, manufacture, and test a novel arthrosopic NIRS fiber probe

• To investigate the potential of arthroscopic NIRS for evaluation of cartilage and subchondral bone properties in equine jointsin vivo

5 MATERIALS AND METHODS

The samples for studiesI-IIIwere acquired from a slaughterhouse in Utrecht, Nether- lands and, thus, no ethical permissions were required. The permission to conduct study IV was obtained from the local Ethics Committee for Animal Experiments in compliance with the Institutional Guidelines on the Use of Laboratory Animals, Utrecht, Netherlands (Permission: DEC 2014.III.11.098). Due to the large number of spectral and reference measurements required for the studies, multiple measure- ments were acquired per joint to ensure reliable model training. A summary of the samples used in this thesis is presented in Table 5.1.

Five equine fetlock joints were utilized in studiesItoIII. In studyIV, anin vivo experiment was conducted in Utrecht University, in which two artificial defects were created on both knees of Shetland ponies (N= 7), and filled with four experimental repairs (fibrin glue,GelMA cap,GelMA, andreinforced GelMA) [126]. After 12 months follow-up, the ponies were sacrificed and their knees examined in arthroscopy to in- vestigate the health of osteochondral tissues surrounding the repair sites. Samples including the repair site and interjacent tissue were extracted and subjected for ref- erence analyses. Additionally, similar samples were extracted from control ponies (N= 3).

Table 5.1:Summary of materials and methods utilized in studiesI-IV Study Joint Number of Measurement Methods

joints points

I Equine fetlock N= 5 n= 43 Arthroscope, OCT, ultrasound, Artscan 200, mechanical testing II Equine fetlock N= 5 n= 869 in vitroNIRS,

mechanical testing III Equine fetlock N= 5 n= 530 in vitroNIRS, OCT,

quantitative microscopy IV Equine stifle N= 20 n= 236 in vivoNIRS,µCT,

mechanical testing