Development of quantitative cest MRI methods with clinical applications

DEVELOPMENT OF QUANTITATIVE CEST MRI METHODS WITH CLINICAL

APPLICATIONS

AYAN DEBNATH

CENTRE FOR BIOMEDICAL ENGINEERING

INDIAN INSTITUTE OF TECHNOLOGY DELHI

OCTOBER 2019

© Indian Institute of Technology Delhi (IITD), New Delhi, 2019

DEVELOPMENT OF QUANTITATIVE CEST MRI METHODS WITH CLINICAL

APPLICATIONS

AYAN DEBNATH

CENTRE FOR BIOMEDICAL ENGINEERING Submitted

in fulfilment of the requirements of the degree of Doctor of Philosophy to the

INDIAN INSTITUTE OF TECHNOLOGY DELHI

OCTOBER 2019

i

CERTIFICATE

This is to certify that the thesis entitled “DEVELOPMENT OF CEST MRI METHODS

Institute of Technology Delhi, for the award of “Doctor of Philosophy” in Centre for Biomedical Engineering is a record of the bona fide research work carried out by him in close supervision by me. He has fulfilled the requirements for submission of this thesis. This work has not been submitted elsewhere for degree.

……….

Dr. Anup Singh Assistant Professor Centre for Biomedical Engineering Indian Institute of Technology Delhi New Delhi – 110016 India

Date: 23

October 2019

Place: New Delhi, India

iii

ACKNOWLEDGEMENTS

I would like to thank my supervisor Dr. Anup Singh for providing me the opportunity to carry out research work with him. He introduced me to the field of CEST MRI and mentored me to learn the subject not only by skill and hard work but also by heart and soul. He helped me to evolve myself by improving my research acumen, alacrity, skill and experience. He motivates me all the time to work with undying energy whenever I got sidetracked. When others doubted me or even when I had self-doubt, he is the only person who firmly believed in me. His immense support is remarkable and provided me enough strength to carry on my research surpassing obstacles with right clarity of thoughts and visions. Apart from being my academic advisor, he is Guru and a friend to me. We had spent good quality of time through innumerable interactions over the years.

I would like to express my sincere regards to Dr. RK Gupta, head of the Department of Radiology, Fortis hospital. I am very much grateful to him for providing us with clinical data and allowing us access to MR scanners. His immense enthusiasm inspires me. His idea and advice shaped up my research work.

I would like to thank my SRC committee members Prof Veena Koul and Dr Kedar Khare for their consideration of my research and sharing their valuable time in meetings. I would like to show my sincere gratitude to Dr. Neetu Singh for having intense confidence in me and motivating me. I am thankful to Dr. Amit Mehndiratta for his kind support, sharing humor and fun.

I am highly indebted to Fulbright Doctoral Research Fellowship for providing me golden lifetime opportunity to involve in research work at world’s top-notch universities Massachusetts General hospital, Harvard Medical School and University of Pennsylvania (UPenn). I consider myself to very fortunate to be research fellow at Dr. Ravinder Reddy’s lab at UPenn. He was my joint supervisor at the United States. His lab provides the best working environment to explore and learn, to have discussions with fellow lab mates and allowed to get access to MR ultra-high field state-of-the-art scanners. I have worked closely with Dr. Hari Hariharan at UPenn. He is genuinely helpful and extended his help to me whenever I got stuck with research problems. I am not exaggerating in stating that he is no less than walking library in the CEST domain. His way of making me things understandable is noteworthy and I learned a lot from him through long hours of several discussions. He is a great teacher with humble and friendly personality. I do have dire admiration for him. I am also grateful to Dr. Ravi Nanga who taught me to carry out MR scanning at 7T scanner and Dr. Puneet Bagga for sharing friendship bond. I would also like to thank to the entire Fulbright team, special mention to Pratibha Nair and Sudarshan Das.

I am very thankful to the entire team of MedImg with whom I spent most of my time. I am

grateful for their companionship, prompt help, support and for providing a pleasurable and friendly

working atmosphere. I express my gratitude to peer lab mates Dr. Anirban, Dr. Snekha, Esha,

Dinil, Rupsa, Rafeek, Sameer, Dharmesh, Archana, Sandeep, Virender, Neha, Banasmitha, Umang

iv

and Manish. I sincerely thank Manish for sharing his knowledge and helping me every time I require, our moto lies in “learn together, grow together”. I want to thank Dr. Mamata Gupta for editing manuscripts and for supporting me mentally ample number of times. I wish to thank the entire faculties of my department, staff and office members.

I am very thankful to my constant companion Aditya at hostel. He is always willing to help me in various ways and supported me in several scenarios. He always listened to my problem with sympathetic ear and dragged me out of any hard situations. I wish to express my gratitude to Snehasis who is my friend, philosopher and guide to me in real life. We share a bond of unconditional love and affection. He is more than a friend of mine, I look up to him as my elder brother. I would like to thank Shubhrima for providing me right directions in research whenever I felt lost, editing my manuscript and being very good friend of mine. I am incredibly thankful to Shushobhit for his constant support and constructive advices. My eternal gratitude to many friends of mine for being in my life.

I would like to thank Dr. Manali who stood by me through thick and thin. Whenever I felt stressed or morally down, I used to vent out everything to her and she is the only person who can console me at this situation patiently through long discussions. She always motivates me to strive for the best, constantly encouraged and pushed me to work harder to overcome difficult scenarios.

She firmly believes in my utmost potential that I can do wonders. I am very fortunate for the immense love she bestowed on me.

Last but not the least, I would like to share my unconditional love for my parents and brother who are the driving force of my life. My father always motivates and inspires me to do to the best of my effort and to be consistent. My mother is the one and only one person who blindly believes in my potential and thinks that her son can achieve everything with his wisdom. My brother always motivates me to dream big. I am indebted for their endless support, constant encouragement, faith and unflagging love. Thank you for being always there.

………

Ayan Debnath

v

ABSTRACT

Chemical exchange saturation transfer (CEST) Magnetic Resonance Imaging (MRI) is a novel non-invasive molecular imaging technique to detect metabolites of very low concentration at high spatial resolution. CEST MRI has shown promising potential for disease diagnosis. Despite its wide range of applications, CEST MRI suffers from several limitations such as long acquisition scan time, field inhomogeneities, overlap of CEST effects from different metabolites. The purpose of the dissertation work was to optimize parameters to carry out CEST experiment in feasible scan time for clinical settings along with development of methods for better quantitation of CEST contrast. Another objective of the thesis was to implement the proposed methods in clinical applications.

For investigating the effect of offset-frequency step-size or sampling frequency on CEST contrast for different CEST metabolites at various field strength, CEST-weighted images were acquired at different step-size and compared. The lower step-size provided good quality map and the maps becomes coarser with the increase in step-size. The error in CEST contrast computation increased with increase in step-size. Different interpolation methods to correct for B

inhomogeneity were compared and the optimal method was investigated to obtain accurate CEST contrast. Results showed that in vivo CEST data should be acquired at step-size between 0.2 ppm to 0.3 ppm for clinical settings along with 2

or 3

degree polynomial interpolation for B

inhomogeneity correction.

The current study evaluated the feasibility of Creatine-weighted (Cr-w) CEST MRI in

human brain at ultra-high field strength MR scanner. To the best of our knowledge, this is the first

study on Cr-w CEST in human brain. The Cr-CEST experiment should be carried out at optimal

saturation parameters of 1.45 µT and 2s. We developed new Z-spectral fitting model for better

computation of CEST contrast. In the new approach, the semi-solid macromolecular magnetization

transfer (MT) was removed from the Z-spectra followed by fitting using super-position of

Lorentzian functions (LS). The width of different components of Z-spectra is narrower using

proposed than conventional LS. Monte-Carlo simulations showed that proposed approach has

better noise stability for the data used in the current study.

vi

The current study also attempted to remove confounding effect from semi-solid macromolecular magnetization transfer (MT) to accurately compute Glutamate-weighted CEST contrast (GluCEST). The study compared different lineshapes such as Lorentzian, Gaussian, super-Lorentzian and 6

degree polynomial function to model MT components. The results, which involved data from healthy human volunteers at 7T and a rat brain tumor data on 9.4T, support the use of Lorentzian fitting to remove MT contribution for improved computation of GluCEST. This approach increased the specificity of GluCEST.

Amide proton transfer-weighted (APT-w) CEST MRI has also been explored to differentiate between neo-plastic and infective mass lesions, as well as intra-cranial mass lesions (ICMLs) using different contrast normalization, ROIs selection and histogram analysis. The study showed that the APT-w contrast (normalized with reference signal at negative offset-frequency and APT-w contrast in normal-appearing-white-matter) corresponding to contrast enhancing or active lesion region provided a large number of histogram parameters to significantly (p<0.05) differentiate among different groups of ICMLs. APT-MRI should be combined with other MRI techniques to further improve the differential diagnosis of ICMLs.

In a nutshell, the thesis presented a comprehensive study of method developments for

improving CEST contrast computation, optimization of offset-frequency step-size and

interpolation methods for B

inhomogeneity correction and some clinical applications of proposed

methodologies. The current research work would be beneficial for carrying out further CEST MRI

related studies and improve diagnosis.

vii

सार

रासायनिक निनिमय संतृनि हसतांतरण ( सी . ई . एस . टी .) च ंबकीय अि िाद इमेन ंग ( एम . आर . आई .) उच्च सथानिक निभेदि पर बहुत कम मात्रा के चयापचयों का पता लगािे के नलए एक अनितीय गैर आक्रामक आणनिक इमेन ंग तकिीक है। सी . ई . एस . टी . एम . आर . आई . िे रोग निदाि के नलए आशा िक क्षमता नदखाई है। अि प्रयोगों की अपिी

निसतृत श्ृंखला के बाि ूद, सी.ई.एस.टी. एम.आर.आई. लंबे अनिग्रहण सकैि समय, क्षेत्र असमांगता, निनभन्ि

चयापचयों से सी . ई . एस . टी . प्रभाि के ओिरलैप आनद से प्रभानित है। शोि प्रबंि कायय का उद्देश्य सी . ई . एस . टी . कंट्रासट के बेहतर पररमाणीकरण के नलए निनियों के निकास के साथ-साथ िैदानिक सेनटंग्स के नलए व्यिहायय सकैि

समय में सी.एस.टी. प्रयोग करिे के नलए मािको को अि कूनलत करिा था। थीनसस का एक अन्य उद्देश्य िैदानिक अि प्रयोगों में प्रसतानित निनियों को लागू करिा था।

ऑफसेट आिृनि कदम आकार या निनभन्ि क्षेत्र ताकत पर निनभन्ि सी . ई . एस . टी . चयापचयों के नलए सी . ई . एस . टी . कंट्रासट पर िमूिा आिृनि के प्रभाि की ांच के नलए, सी.ई.एस.टी. भाररत छनियों अलग कदम आकार पर अनिग्रहण नकया गया और त लिा की गई। कम कदम आकार िे अच्छी ग णििा के िक्शे प्रदाि नकये है और िक्शे

कदम आकार में िृनि के साथ मोटे हो ाते है। कदम आकार में िृनि के साथ सी.ई.एस.टी. कंट्रासट गणिा में त्र नट बढ़ गई। बी 0 असमांगता को सही करिे के नलए निनभन्ि अंतिेशि निनियों की त लिा की गई और इष्टतम निनि

सटीक सी.ई.एस.टी कंट्रासट प्राि करिे के नलए ांच की गई। पररणामों से पता चला है िैदानिक सेनटंग्स के नलए नक इि नििो सी . ई . एस . टी . डेटा 0.2 पीपीएम से 0.3 पीपीएम के बीच कदम आकार में प्राि नकया ािा चानहए और बी0 असमांगता स िार के नलए 2 या 3 नडग्री बहुपद अंतम यखता का प्रयोग नकया ािा चानहए।

ितयमाि अध्ययि िे अल्ट्ट्रा उच्च क्षेत्र शनि एमआर सकैिर पर मािि मनसतष्क में नक्रएनटनिि भाररत सी.ई.एस.टी.

एम.आर.आई. की व्यिहाययता का मूल्ट्यांकि नकया। हमारे ज्ञाि के अि सार, यह मािि मनसतष्क में नक्रएनटनिि भाररत सी.ई.एस.टी. पर पहला अध्ययि है। नक्रएनटनिि भाररत सी.ई.एस.टी. प्रयोग 1.45 µT और 2s के इष्टतम संतृनि

मापदंडों पर नकया ािा चानहए। हमिे सी.ई.एस.टी. कंट्रासट की बेहतर गणिा के नलए िए ेड-सपेक्ट्रल नफनटंग मॉडल निकनसत नकए हैं। िए दृनष्टकोण में, अिय ठोस मैक्रोआनविक च ंबकि हसतांतरण (एम.टी.) को ेड-सपेक्ट्रा

से हटा नदया गया न सके बाद लोरेंट्ऩियि फंक्शि की स पर- पोन शि का उपयोग करके नफनटंग की गई। पारंपररक लोरेंट्ऩियि फंक्शि की त लिा में प्रसतानित का उपयोग करके ेड-सपेक्ट्रा के निनभन्ि घटकों की चौडाई संकरी है।

मोंटे - कालो नसम लेशि से पता चला है नक प्रसतानित दृनष्टकोण ितयमाि अध्ययि में इसतेमाल नकया डेटा के नलए बेहतर शोर नसथरता प्रदाि करता है।

ितयमाि अध्ययि में अिय ठोस मैक्रोआनविक च ंबकि हसतांतरण ( एम . टी .) के प्रभाि को दूर करिे के नलए सही

ग्लूटामेट भाररत सी.ई.एस.टी. कंट्रासट की गणिा करिे का प्रयास भी नकया है। अध्ययि में च ंबकि हसतांतरण मॉडल

के घटकों को बिािे के नलये निनभन्ि लाइिशेप ैसे लोरेंट्ऩियि, गाउनसयि , स पर लोरेंट्ऩियि और 6 नडग्री बहुपद

viii

फंक्शि की त लिा की गयी। िो पररणाम

मनसतष्क ट्यूमर डेटा से प्राि हुए,

िो ग्लूटामेट सी.ई.एस.टी. से च ंबकि हसतांतरण घटको को हटािे के नलए लोरेंट्ऩियि फंक्शि का बेहतर समथयि

करते है। यह दृनष्टकोण ग्लूटामेट भाररत सी.ई.एस.टी. की निनशष्टता में िृनि करता है।

एमाइड प्रोटॉि ट्रांसफर-िेटेड (एपी.टी.डब्लल्ट्यू.) सी.ई.एस.टी. एम.आर.आई. को िि-प्लानसटक और संक्रामक घािों के बीच अंतर करिे और अंतर-कपाल द्रव्यमाि घािों (आई.सी.एम.एल) को पता लगािे के नलये निनभन्ि

निपरीत सामान्यीकरण, आर. ओ. आई. चयि और नहसटोग्राम निश्लेषण का उपयोग नकया गया है। अध्ययि से पता

चला है नक निपरीत बढ़ािे या सनक्रय घाि क्षेत्र ए . पी . टी -डब्लल्ट्यू. कंट्रासट के ( िकारात्मक ऑफसेट आिृनि पर संदभय संकेत के साथ सामान्यीकृत और सामान्य-प्रदनशयत-सफेद मैटर में ए.पी.टी.-डब्लल्ट्यू. कंट्रासट) अि रूप नहसटोग्राम की

एक बडी संख्या प्राि हुई ो आई . सी . एम . एल . के निनभन्ि समूहों के बीच अंतर (p<0.05) करिे में महत्िपूणय है।

ए.पी.टी. एम.आर.आई. को आई.सी.सी.एल. के निभेदक निदाि में और स िार करिे के नलए अन्य एमआरआई तकिीकों के साथ ोडा ािा चानहए।

संक्षेप में, थीनसस में सी.ई.एस.टी. कंट्रासट गणिा स िार के नलए निनि के निकास, ऑफसेट आिृनि कदम आकार और बी 0 असमांगता स िार और क छ िैदानिक अि प्रयोगों के नलए प्रक्षेपक तरीकों का एक व्यापक अध्ययि प्रसत त नकया गया। ितयमाि अि संिाि कायय सी.ई.एस.टी. एम.आर.आई. संबंनित अध्ययि को आगे ले ािे और निदाि

में स िार के नलए फायदेमंद होगा।

ix

CONTENTS

CERTIFICATE i

ACKNOWLEDGEMENT iii

ABSTRACT v

CHAPTER 1 1

INTRODUCTION, LITERATURE REVIEW AND BRIEF OUTLINE OF THESIS 1

1.1. History and basics of MRI 2

1.2. CEST MRI 4

1.2.1. Brief history of CEST MRI 4

1.2.2. Basic principles of CEST MRI 5

1.2.3. Pulse sequence design of CEST MRI 7

1.2.4. Factors influencing CEST MRI 11

1.2.4.1. Effect of concentration of metabolites on CEST contrast 11

1.2.4.2. Effect of exchange rates on CEST contrast 12

1.2.4.3. Effect of pH on CEST contrast 13

1.2.4.4. Effect of field strength on CEST contrast 14

1.2.4.5. Effect of field inhomogeneities on CEST contrast 15

1.2.4.6. Effect of pulse shape, number of pulses on CEST contrast 16

1.2.4.7. Effect of data acquisition strategy on CEST contrast 16

1.2.5. Computation of CEST contrast: Key developments 17

1.2.6. Clinical applications of CEST MRI 24

1.2.6.1. Amide Proton Transfer weighted imaging 24

1.2.6.1.1. APT-w MRI in stroke 24

1.2.6.1.2. Amide Proton Transfer weighted imaging in tumor detection and grading 25 1.2.6.1.3. Amide Proton Transfer weighted imaging in various pathologies 25

1.2.6.2. Creatine weighted CEST MRI 26

1.2.6.3. Glutamate weighted CEST MRI 27

1.2.6.3.1. Pilot study on GluCEST MRI 27

1.2.6.3.2. GluCEST in Alzheimer’s disease 28

1.2.6.3.3. GluCEST in brain tumors 28

1.2.6.3.4. GluCEST in various other neurological disorders 28

1.3. Challenges and motivations 29

1.5. Objectives 32

1.6. Thesis outline 32

x

CHAPTER 2 35

EVALUATING EFFECT OF SATURATION OFFSET-FREQUENCY STEP-SIZE AND INTERPOLATION METHODS ON CEST CONTRAST IN HUMAN BRAIN AT 3T AND

7T MR SCANNERS 35

2.1. Introduction 37

2.2. Materials and methods 38

2.2.1. Simulation 38

2.2.1.1. Simulation on APT-w CEST MRI 38

2.2.1.2. Simulation on Glu-w CEST MRI 39

2.2.1.3. Simulation on Cr-w CEST MRI 39

2.2.2. Human study 40

2.2.2.1. GluCEST MRI at 7T 40

2.2.2.2. Human study of APT-w CEST MRI at 3T 41

2.2.3. Post processing 42

2.2.4. B0 inhomogeneity correction 42

2.2.5. B1 inhomogeneity correction 43

2.2.6. Statistical analysis 44

2.3. Results 44

2.3.1. Simulations 44

2.3.1.1. Simulation on APT-w CEST MRI 44

2.3.1.2. Simulation on Glu-w CEST MRI 45

2.3.1.3. Simulation on Cr-w CEST MRI 45

2.3.2. Human study 49

2.3.2.1. GluCEST MRI at 7T 49

2.3.2.2. APT-w MRI at 3T 54

2.4. Discussion 58

2.5. Conclusion 61

CHAPTER 3 63

EVALUATING FEASIBILITY OF CREATINE-WEIGHTED CEST MRI IN HUMAN

BRAIN AT 7T USING Z-SPECTRAL FITTING APPROACH 63

3.1. Introduction 65

3.2. Materials and methods 67

3.2.1. CEST MRI pulse sequence 67

xi

3.2.2. Phantom Study 67

3.2.2.1. Phantom preparation 67

3.2.2.2. Acquisition protocol for phantom MR imaging 68

3.2.3. Simulation 68

3.2.4. In Vivo human Brain MRI 70

3.2.5. CEST contrast computation 71

3.2.6. Data analysis / post processing 72

3.2.7. Monte-Carlo simulation 74

3.2.8. B0 inhomogeneity correction for CESTasym contrast 75

3.2.9. B1 inhomogeneity correction 75

3.2.10. Statistical analysis 75

3.3. Results 76

3.3.1. Phantom 76

3.3.2. Simulation 77

3.3.3. In vivo Human study 79

3.4. Discussion 94

3.5. Conclusion 98

CHAPTER 4 99

GLUTAMATE-WEIGHTED CEST CONTRAST AFTER REMOVAL OF

MAGNETIZATION TRANSFER EFFECT IN HUMAN BRAIN AND RAT BRAIN WITH

TUMOR 99

4.1. Introduction 101

4.2. Materials and methods 104

4.2.1. Human studies 104

4.2.1.1. subject recruitment 104

4.2.1.2. Acquisition protocol for human scan 104

4.2.2. Animal studies 105

4.2.2.1. Animal preparation and experiment 105

4.2.2.2. Imaging protocol for in vivo animal study 105

4.2.3. B0 and B1 inhomogeneity correction 105

4.2.4. MT modelling by different lineshapes: fitting functions 106

4.2.5. Post processing 107

4.2.6. Statistical analysis 108

4.3. Results 108

4.3.1. MT modelling 108

4.3.2. CEST asymmetry analysis 112

4.3.3. GluCEST maps before and after MT removal 113

xii

4.3.4. Interpolation for super-Lorentzian lineshape 117

4.3.5. Rat tumor model 118

4.4. Discussion 119

4.5. Conclusion 122

CHAPTER 5 123

EVALUATING THE ROLE OF AMIDE-PROTON-TRANSFER (APT) WEIGHTED CONTRAST, OPTIMIZED FOR NORMALIZATION AND REGIONS OF INTEREST SELECTION, IN DIFFERENTIATION OF NEOPLASTIC AND INFECTIVE MASS

LESIONS ON 3T MRI 123

5.1. Introduction 125

5.2. Materials and methods 126

5.2.1. Patient recruitment 126

5.2.2. MR imaging sequence and protocol 126

5.2.3. Theory 128

5.2.4. Image processing and analysis 129

5.2.5. ROIs selection and analysis 129

5.2.6. Histogram analysis 130

5.2.7. Statistical analysis 130

5.3. Results 131

5.3.1. Pathological diagnosis 131

5.3.2. APT-w contrast measurement by different normalizations 131

5.3.3. Statistical analysis 139

5.3.3.1. Shapiro-Wilk’s: Test for normality 139

5.3.3.2. T-Test 139

5.3.3.3. ANOVA with post-hoc test 141

5.3.3.4. ROC analysis 141

5.4. Discussion 146

5.5. Conclusion 149

CHAPTER 6 151

CONCLUSION AND FUTURE DIRECTIONS 151

6.1. Summary with major findings of the thesis 152

xiii

6.2. Additional work with future directions 155

6.2.1. Development of CEST pulse sequence and phantom design for testing 156 6.2.2. Development of Z-spectral analysis using probabilistic approach and comparison of different Z-spectral

methods 157

References 167

List of Publications 181

Journals 181

International conference proceedings 181

National conference proceedings 183

Special achievement / Awards 184

Fellowship 184

Workshop attended 184

Curriculum Vitae 185

xv Abbreviations

RF – Radio-frequency FID – Free Induction Decay

B

- Root mean square of B

(RF amplitude) DS - Direct Saturation

MT - Magnetization Transfer

CEST - Chemical Exchange saturation transfer rNOE – relayed nuclear Overhausser effect CEST

– CEST asymmetry analysis Cr-w CEST – Creatine-weighted CEST GluCEST – Glutamate-weighted CEST APT-w – Amide proton transfer-weighted SAR – Specific Absorption Rate

CV – Coefficient of Variation Corr - Coefficient of Correlation

nMSE – normalized Mean of Square Errors R

– Goodness of fit

ARPE – Absolute Relative Percentage Errors AUC - Area Under Curve

ICMLs – Intra-Cranial Mass Lesions LGG – Low Grade Glioma

HGG – High Grade Glioma

ROC – Receiver Operating Characteristics

xvi

AROC - Area under curve for ROC analysis TR – Repetition time

TE – Echo time

T

– Longitudinal relaxation time or spin-lattice relaxation time T

– Transverse relaxation time or spin-spin relaxation time WM – White Matter

GM – Gray Matter

NAWM – Normal Appearing white Matter CNR - Contrast to Noise Ratio

GRE – Gradient Recalled Echo SE – Spin Echo

GRASE – Gradient and Spin Echo FLASH – Fast Low Angle shot

MP-RAGE - Magnetization Prepared Rapid Gradient Echo

xvii

List of Figures Figure

No.

Figure Description Page

No.

Figure 1. 1: Two pool system with one solute pool having exchangeable protons and other free water pool. (a) protons of solute pool (shown in red-color) and bulk water protons. (b) Protons of pools get exchanged after some time, solute pool proton moves to bulk water pool and vice-versa. ... 5 Figure 1. 2: Magnetization transfer between pool A and pool B due to chemical exchange phenomenon.

(a) Both the pools have two energy levels with spins having different angular momentum. (b) When RF pulse is applied at a saturating frequency of the pool B, few spins moves from lower energy level to higher energy levels of pool B. After Rf pulse is applied, the colour of the pool B is changed to red to show that the pool B gets saturated (c) Transfer of magnetization of spins from one pool to another. Pool B is getting continuously saturated due to application of RF pulse. ... 6 Figure 1. 3: A continuous wave (CW) CEST saturation preparation pulse followed by gradient echo readout (GRE). ... 8 Figure 1. 4: Pulse sequence diagram for CEST imaging. A train of preparation saturation pulses was applied followed by lipid suppression and 3D GRASE readout. Adapted from source: Zhu et al MRM, 2010. DOI: 10.1002/mrm.22546 ... 10 Figure 1. 5: Pulse sequence diagram of a 3D CEST imaging with FLASH readout. Adapted from Cai et al¹², Nat Med, DOI: 10.1038/nm.2615 ... 10 Figure 1. 6: Two pool model to compute CEST contrast using asymmetry analysis. One pool

corresponding to bulk water is considered for upper row. (a) water spectrum in NMR. (b) Z-spectrum of water pool. (c) asymmetry analysis of water peak or direct saturation provides zero signal. It shows that direct saturation is symmetric about symmetric about water resonating frequency. Two pools are considered corresponding to bulk water and a solute pool for creatine for lower row. (d) NMR spectrum shows a very high peak for water and a very small peak for metabolites with low concentration. (e) Z- spectrum shows two dips for water and creatine providing CEST effect. (f) Since the Z-spectrum with two pool is asymmetric, thus asymmetric analysis provide CEST contrast... 19 Figure 1. 7: (a) anatomy image. (b) B0 map. B0 inhomogeneity various across a slice. ... 20 Figure 1. 8: Z-spectra and CEST asymmetry analysis with and without B0 inhomogeneity. Due to B0

inhomogeneity, (a) each data points in Z-spectra gets shifted resulting in (b) erroneous CEST asymmetry or CEST contrast. ... 20 Figure 1. 9: CESTasy is confounded by several factors. (a) the magnetization signal before RF irradiation pulse. (b) There is a signal drop after RF pulse is applied. This signal drop is supposed to be due to only CEST contrast, but in reality, it is mixture of several effects from direct saturation (DS), magnetization transfer (MT) and CEST effects... 21 Figure 1. 10: Flowchart of workflow or pipeline of thesis ... 34

Figure 2. 1: Comparison of different interpolation methods for CEST-weighted signal intensity (SI) over the frequency range of +1.8 ppm to +4.2 ppm. Fitting of SI corresponding to an ROI in WM (a1-a6) with step-size of 0.1 ppm. Fitting of SI corresponding to voxel within that ROI with step-size of 0.1 ppm (b1- b6), 0.2 ppm (c1-c6), 0.3 ppm (d1-d6), 0.4 ppm (e1-e6) and 0.5 ppm (f1-f6). The columns 1-3 show interpolation methods using polynomial degree 1, 2 and 3 respectively. The columns 4-6 show

xviii

interpolation methods using cubic, spline and smoothingspline respectively. R² represents goodness-of-fit.

... 50 Figure 2. 2: A representative case of B0 and B1 corrected GluCESTNeg maps of a healthy volunteer for different step-size and interpolation methods. The rows are for different sampling offset-frequency step- sizes of 0.1 ppm (a1-a6), 0.2 ppm (b1-b6), 0.3 ppm (c1-c6), 0.4 ppm (d1-d6) and 0.5 ppm (e1-e6). The columns 1-3 show interpolation methods using polynomial degree 1, 2 and 3 respectively. The columns 4- 6 show interpolation methods using cubic, spline and smoothingspline respectively. ... 51 Figure 2. 3: A representative case of B0 and B1 corrected GluCESTM0 maps of a healthy volunteer for different step-size and interpolation methods. The rows are for different sampling offset-frequency step- sizes of 0.1 ppm (a1-a6), 0.2 ppm (b1-b6), 0.3 ppm (c1-c6), 0.4 ppm (d1-d6) and 0.5 ppm (e1-e6). The columns 1-3 show interpolation methods using polynomial degree 1, 2 and 3 respectively. The columns 4- 6 show interpolation methods using cubic, spline and smoothingspline respectively. ... 52 Figure 2. 4: Corr and nMSE measurements for GluCESTNeg contrast averaged over all volunteers for WM mask. The columns 1-3 show interpolation methods using polynomial degree 1, 2 and 3 respectively.

The columns 4-6 show interpolation methods using cubic, spline and smoothingspline respectively. ... 53 Figure 2. 5: Corr and nMSE measurements for GluCESTNeg contrast over GM and WM (column 1-2) and GluCESTM0 contrast over GM and WM mask (column 3-4) using 2^nd order polynomial interpolation method for B0 inhomogeneity correction. The Corr decreases and nMSE increases with increase in step- size. ... 55 Figure 2. 6: B0 inhomogeneity corrected APT-w maps of representative LGG (first row) and HGG (second row) using different interpolation methods. The red arrow represents tumor regions. Since cubic and spline functions are based on piecewise polynomial functions, thus cubic and spline functions are picking up noise in partial volume regions and doing over fitting. Polynomial functions with 2^nd and 3^rd order degree are providing better interpolation methods. ... 56 Figure 2. 7: Quantitative comparison using box and whisker plots of 𝐴𝑃𝑇𝑁𝑒𝑔_𝑁𝐴𝑊𝑀 contrast between LGG and HGG corresponding to ROI-1 and ROI-2. Polynomial interpolations provide significant

difference (p<0.05) between the groups. ... 57

Figure 3. 1: (a) Z-spectra of Cr (10 mM) and PBS using saturation B1rms of 2.2 µT and duration of 1s. (b) Dependence of Cr-w CESTasy contrast on Cr concentration ([Cr]) using B1rms of 2.2 µT and duration of 1s. (c) dependence of Cr-w CEST contrast on saturation power for a fixed duration of 1s and (d) dependence on saturation duration for a fixed saturation B1rms of 2.2 µT. Phantom data was acquired at temperature of 37±1 ^oC and pH of 7.0. Here, y represents CrCEST (%) and x represents [Cr] in (b), B1rms

in (c) and saturation duration in (d). ... 76 Figure 3. 2: Surface plots demonstrating dependence of Cr-weighted (w) CEST contrast at 1.8 ppm on B1rms and saturation duration using numerical simulations at 7T. (a) CEST contrast from Cr [10mM] only.

(b) CEST effect from combined metabolites [MI, Glu/GABA, APT] except Cr at 1.8 ppm. Arrows point to the three different saturation values and corresponding contrast. Circles show regions corresponding to optimum range of saturation parameters. Dotted ellipse highlights the regions for optimal CrCEST

contrast with less contamination. ... 78 Figure 3. 3: Surface plots demonstrating dependence of CEST contrast at 1.8 ppm from Cr [10mM] on B1rms and saturation duration using numerical simulations at 7T. (a, b) Cr-w CEST with exchange rate of 900 Hz and 1190 Hz respectively. ... 78

xix

Figure 3. 4: Z-spectra (a) and asymmetry plot (b) from an regions of interest (ROI) in gray matter tissue of human brain at 7T with saturation pulse B1rms of 0.7, 1.45, 2.2, 2.9 µT and duration of 2s. Asymmetry plots in (b) use normalization by signal without saturation (M0). ... 79 Figure 3. 5: (a) CEST image without saturation, (b) CEST image at 1.8 ppm, (c) B0 map, (d) B1 map and (e) CESTasy map at 1.8 ppm using normalization by CEST image without saturation (M0). ... 80 Figure 3. 6: Plots show an example of fitting in vivo Z-spectra, from an ROI in gray matter tissue, for B1rms of 1.45 µT and duration of 2 s using Model-1. (a) Original Z-spectra (Z1). (b) Scaled Z-spectra (Z2).

(c) 5-pool fitting of scaled Z-spectra (Z2) along with various fitted components DS, MT, rNOE-w, Cr-w CEST and CEST@3.5 ppm contrast. R² = 0.99 and residuals errors are under 2%. (d) Individual CEST and rNOE components. Fitting is done over range of [-40 to +40] ppm and shown for range of [-20 to +20] ppm for better visualization. ... 81 Figure 3. 7: Plots show an example of fitting in vivo Z-spectra, from ROI in gray matter tissue, for B1rms

of 1.45 µT and duration of 2 s using Model-2. (a) Original Z-spectra (Z1). (b) Scaled Z-spectra (Z2) along with fitted MT component and Z-spectra after removal of MT component (Z3). (c) 4 pools fitting of Z3 along with various fitted components like DS, rNOE, Cr-w CEST and CEST@3.5 ppm contrast. R²

= 0.99 and residuals errors are under 2%. (d) Individual CEST and rNOE components. Fitting of Z2 is done over range of [-40 to +40] ppm and shown for range of [-20 to +20] ppm for better visualization. .. 82 Figure 3. 8: Fitting of in vivo Z-spectra from gray matter tissue using Model-2 at durations of 2s and B1rms

of 0.7 µT (a), 1.45 µT (b), 2.2 µT (c). Residuals errors of fitting are under 2%. rNOE decreases with increase in B1rms. Cr-w CEST and CEST@3.5 ppm initially increases and then decreases with B1rms. ... 85 Figure 3. 9: Amplitude and width maps of various components of Z-spectra from a representative healthy human brain using Model-1 (row-1, 3) and Model-2 (row-2, 4). Amplitude maps of DS (a, f), MT (b, g), rNOE (c, h), Cr-w CEST (d, i), CEST@3.5 ppm (e, j). Width maps of DS (k, p), MT (l, q), rNOE (m, r), Cr-w CEST (n, s), CEST@3.5 ppm (o, t). The scales for width maps are in ppm unit. White arrows point to representative subarachnoid space of brain. Images are cropped. ... 86 Figure 3. 10: Plots show the sensitivity of Cr-w CEST contrast computed using fitting approach to shift in resonance frequency (delB0). Z-spectrum from an ROI in gray matter tissue was selected followed by fitting. Simulated Z-spectra were generated by varying the center, corresponding to direct saturation of free water obtained using fitting, from -0.5 to 0.5 ppm. These simulated Z-spectra were fitted using model-2 for computing Cr-w CEST as well as used for computing asymmetry contrast. ... 87 Figure 3. 11: Fitting of Z-spectra using Model-2 at B1rms of 1.45 µT and duration of 2 s using multi-pool models. (a) Fitting using 5 pool model (DS pool at 0ppm, MT pool at -2.4 ppm, rNOE pool at -3.5 ppm, Cr-w CEST pool at 1.8 ppm and CEST@3.5ppm. (b) Fitting considering 6 pools, where first 5 pools are similar to that of (a) and the 6^th pool (CEST@3ppm) is considered at 3 ppm. (c) 6^th pool (CEST@1ppm) is considered at 1 ppm. (d) 6^th pool (rNOE-2) is considered at -1.6 ppm. Residuals of fittings were under 2% and  = ~ 0.05 for all cases. ... 90 Figure 3. 12: Fitting of Z-spectra using Model-2 at B1rms of 0.7 µT and duration of 2 s using multi-pool models. (a) Fitting using 5 pool model (DS pool at 0ppm, MT pool at -2.4 ppm, rNOE pool at -3.5 ppm, Cr-w CEST pool at 1.8 ppm and CEST@3.5ppm. (b) Fitting considering 6 pools, where first 5 pools are similar to that of (a) and the 6^th pool (CEST@3ppm) is considered at 3 ppm. (c) 6^th pool (CEST@1ppm) is considered at 1 ppm. (d) 6^th pool (rNOE-2) is considered at -1.6 ppm. Residuals of fittings were under 2% and  = ~0.06 for all cases. ... 91 Figure 3. 13: Fitting of Z-spectra using Model-2 at B1rms of 2.1 µT and duration of 2 s using multi-pool models. (a) Fitting using 5 pool model (DS pool at 0ppm, MT pool at -2.4 ppm, rNOE pool at -3.5ppm, Cr-w CEST pool at 1.8 ppm and CEST@3.5ppm. (b) Fitting considering 6 pools, where first 5 pools are similar to that of (a) and the 6^th pool (CEST@3ppm) is considered at 3 ppm. (c) 6^th pool (CEST@1ppm)

xx

is considered at 1 ppm. (d) 6^th pool (rNOE-2) is considered at -1.6 ppm. Residuals of fittings were under 1% and  = ~ 0.065 for all cases. ... 92 Figure 3. 14: Partial fitting of in vivo Z-spectra at far-off resonance frequency range ±40 ppm to ±12 ppm using Lorentzian, Gaussian and super-Lorentzian lineshapes. (a, b, c) corresponds to fitting of Z-spectra acquired at B1rms of 0.7 µT. (d, e, f) corresponds to fitting of Z-spectra acquired at B1rms of 1.4 µT. (g, h, i) corresponds to fitting of Z-spectra acquired at B1rms of 2.1 µT. Column 1-3 corresponds to fitting using Lorentzian, Gaussian and super-Lorentzian lineshapes respectively. Average residual error () is least for Lorentzian compared with other lineshapes. Lorentzian provided better approximation to broad MT component of Z-spectra for the current set of data. The interpolation of super-Lorentzian lineshape produces a hump at -2.4 ppm. ... 93

Figure 4. 1: Partial Z-spectra fitted with different lineshapes to model MT component and fitted MT component is removed from Z-spectra generated from an ROI drawn over GM. (a) Lorentzian, (b) Gaussian, (c) super-Lorentzian and (d) 6^th degree polynomial function lineshapes were used to fit partial Z-spectra (±100 ppm to ±14 ppm). Fitted MT components were interpolated over the entire frequency range (-100 ppm to +100 ppm) using (e) Lorentzian, (f) Gaussian, (g) spline and (h) 6^th degree polynomial functions. The fitted MT component was subtracted from Z-spectra to get modified Z-spectra without MT component (last row). For better visualization, -70 ppm to +70 ppm data were plotted for first two rows and -10 ppm to +10 ppm data were plotted in last row. ... 109 Figure 4. 2: Partial Z-spectra fitted with different lineshapes to model MT component and fitted MT component is removed from Z-spectra generated from an ROI drawn over WM. (a) Lorentzian, (b) Gaussian, (c) super-Lorentzian and (d) 6^th degree polynomial function lineshapes were used to fit partial Z-spectra (±100 ppm to ±14 ppm). Fitted MT components were interpolated over the entire frequency range (-100 ppm to +100 ppm) using (e) Lorentzian, (f) Gaussian, (f) spline and (h) 6^th degree polynomial functions. The fitted MT component was subtracted from Z-spectra to get modified Z-spectra without MT component (last row). For better visualization, -70 ppm to +70 ppm data were plotted for first two rows and -10 ppm to +10 ppm data were plotted in last row. ... 111 Figure 4. 3: Variation of CESTasy over offset frequencies in (a-d) GM and (e-h) WM before and after MT removal. The MT component was modelled and fitted using (a, e) Lorentzian, (b, f) Gaussian, (c, g) super-Lorentzian with spline interpolation and (d, h) 6^th degree polynomial function lineshapes. CESTasy

increases after removal of MT component. For better visualization, 0 ppm to 5 ppm data were plotted. 112 Figure 4. 4: GluCESTM0 contrast before and after MT removal for different healthy human subjects. The conventional GluCESTM0 contrast before MT removal for all subjects are shown in first column. The GluCESTM0 contrast after MT removal for all subjects using lineshapes to model MT components - (second column) Lorentzian, (third column) Gaussian, (fourth column) super-Lorentzian with spline interpolation and (fifth column) 6^th degree polynomial function lineshapes. There is an elevation in GluCESTM0 contrast after MT removal for all lineshapes. Color bars represents percentage GluCESTM0

contrast. ... 114 Figure 4. 5: GluCESTNeg contrast before and after MT removal for different healthy human subjects. The conventional GluCESTNeg contrast before MT removal for all subjects are shown in first column. The GluCESTNeg contrast after MT removal for all subjects using lineshapes to model MT components - (second column) Lorentzian, (third column) Gaussian, (fourth column) super-Lorentzian with spline interpolation and (fifth column) 6^th degree polynomial functions. The GluCESTNeg contrast decreases after MT removal for all lineshapes. Color bars represents percentage GluCESTNeg contrast. For better

xxi

visualization, the conventional GluCESTNeg contrast (first column) has one scale range and rest of the columns have different scale range. ... 115 Figure 4. 6: (a) Fitting of partial (±100 ppm to ±14 ppm) Z-spectra with super-Lorentzian lineshape to model MT component. Z-spectra was generated from an ROI drawn over GM. (b) Interpolation of fitted Z-spectra (represented by green color) over the entire frequency range (-100 ppm to +100 ppm) using fitted parameters from super-Lorentzian lineshape. The super-Lorentzian lineshape creates a hump or pole at the offset-frequency center of the MT pool. The center was at -1.83 for this Z-spectra. (c) Zoomed version of b. (d) To eliminate the hump, spline interpolation (represented by blue color) has been used over the frequency range (-14 ppm to +14 ppm) instead of using fitted parameters to generate

interpolation by super-Lorentzian lineshape. ... 117 Figure 4. 7: The conventional GluCESTM0 maps of a rat brain with tumor. (a) anatomic brain image of unsaturated signal at 100 ppm (M0) demonstrating ROI-1: tumor region (red circle) and ROI-2: contra- lesion normal appearing region (blue circle); (b) conventional GluCESTM0 map; (c) GluCESTM0 map after removal of MT effect using Lorentzian lineshape; (d) MTR map at 20ppm. ... 118

Figure 5. 1: A representative case of intra-cranial mass lesions (ICMLs). (a) T2-weighted (w), (b) post contrast T1-w image, (c) B0 map and (d) APTNeg map. Active region of lesion show hyper-intense APT-w contrast compared to contra-lateral region and other portion of the brain slice. Necrotic region of the lesion has lower APT-w contrast compared to active-tumor. ... 132 Figure 5. 2: Row 1-3 show MRI images of representative neoplastic mass lesion (low grade glioma (LGG) and high grade glioma (HGG)) and infective mass lesion (tubercular abscess) respectively.

Column 1-9 contain T2-weighted(w), post contrast T1-w (PCT1-w), FLAIR, APT-w images; B0 map, APTM0, APTNeg, APTM0_NAWM and APTNeg_NAWM maps respectively. APTM0_NAWM (Type-3) and

APTNeg_NAWM (Type-4) noramlizations reduces the variability between NAWM and NAGM than APTM0

(Type-1) and APTNeg (Type-2) noramlizations. Red arrows point to lesion region. ... 133 Figure 5. 3: Box and whisker plots showing ROI analysis of APT-weighted(w) contrast in white matter (WM) and grey matter (GM) for Type-1, Type-2, Type-3 and Type-4 normalizations. APT-w contrast for WM is mostly negative and lesser than GM. Type-1, Type-2, Type-3 and Type-4 normalizations

represents APTM0, APTNeg, APTM0_NAWM and APTNeg_NAWM contrast respectively. ... 134 Figure 5. 4: Placement of ROIs on a representative case of LGG. (a) FLAIR image, (b) selection of entire tumor region on that FLAIR image, (c) post-contrast T1-weighted (PC T1-w), (d) selection of active tumor region on that PC T1-w image, (e) PC T1-w image and (f) selection of necrotic tumor region on PC T1-w image. ... 135 Figure 5. 5: Box and whisker plots show quantitative comparison of neoplastic and infective mass lesions for all types of APT-w contrast normalizations corresponding to two ROIs. ROI-1 represents entire lesion and ROI-2 is active-lesion region. Type-1, Type-2, Type-3 and Type-4 normalizations represents APTM0, APTNeg, APTM0_NAWM and APTNeg_NAWM contrast respectively. Different range of scale was used for different normalizations to obtain better representation and visualization of individual plots. ... 137 Figure 5. 6: Box and whisker plots show quantitative comparison of different types APT-w contrast normalizations among different groups of ICMLs for two ROIs. ROI-1 represents entire lesion and ROI-2 is active-lesion region. Type-1, Type-2, Type-3 and Type-4 normalizations represents APTM0, APTNeg, APTM0_NAWM and APTNeg_NAWM contrast respectively. Different range of scale was used for different normalizations to obtain better representation and visualization of individual plots. ... 138

xxii

Figure 5. 7: ROC analysis of mean APTNeg_NAWM-w contrast (Type-4 normalizations) corresponding to ROI-2 for differentiation between neoplastic and infective mass lesion (a) and between low grade glioma and infective mass lesion (b). ... 143

xxiii

List of Tables Table

No.

Table Description Page

No.

Table 2. 1: Simulation result show that Coefficient of variation (CV) of 𝐴𝑃𝑇𝑀0-w contrast increases with the increase in step-size irrespective of interpolation methods. ... 46 Table 2. 2: Simulation result show that normalized-mean-square-error (nMSE) of 𝐴𝑃𝑇𝑀0-w contrast increases with the increase in step-size in all the interpolation methods. ... 46 Table 2. 3: Simulation result show that Coefficient of variation (CV) of 𝐺𝑙𝑢𝐶𝐸𝑆𝑇𝑀0 contrast increases with the increase in step-size irrespective of interpolation methods. ... 47 Table 2. 4: Simulation result show that normalized-mean-square-error (nMSE) of 𝐺𝑙𝑢𝐶𝐸𝑆𝑇𝑀0 contrast increases with the increase in step-size in all the interpolation methods. ... 47 Table 2. 5: Simulation result show that Coefficient of variation (CV) of Cr-w 𝐶𝐸𝑆𝑇𝑀0 contrast

increases with the increase in step-size irrespective of interpolation methods. ... 48 Table 2. 6: Simulation result show that normalized-mean-square-error (nMSE) of Cr-w 𝐶𝐸𝑆𝑇𝑀0 contrast increases with the increase in step-size in all the interpolation methods. ... 48

Table 3. 1: Simulation results on potential contributions to in-vivo human brain Cr-w CEST contrast at 7T. ... 69 Table 3. 2: Fitted amplitude parameter obtained at B1rms of 1.4µT and duration of 2s using model-2 for multiple ROIs on gray matter (GM) and white matter (WM) tissue averaged over all healthy human volunteers ... 84 Table 3. 3: Comparison of Model-2 and Model-1 fitting to Z-spectra data obtained at B1rms of 1.4µT and duration of 2s from gray matter tissue of human brain using Monte-Carlo simulations. ... 88

Table 4. 1: Initial parameters to fit Z-spectra with its lower and upper bounds for MT modelling ... 107 Table 4. 2: Partial fitting of Z-spectra generated from an ROI drawn on GM and WM of a representative healthy human volunteer using different lineshapes to model broad MT spectrum. ... 110 Table 4. 3: GluCEST contrast with and without MT correction ... 116

Table 5. 1: Histogram parameters of Type-3 and Type-4 normalizations, corresponding to ROI-1 and ROI-2, which provide significance difference (p<0.05), using T-test, between neoplastic and infective mass lesions. ... 140 Table 5. 2: Histogram parameters of Type-3 and Type-4 normalizations, corresponding to ROI-1 and ROI-2, which provide significance difference (p<0.05), using One-way ANOVA, between different groups of intra cranial mass lesions. ... 142 Table 5. 3: ROC analysis of Type-4 normalized APT-w contrast corresponding to ROI-2 for

differentiating neoplastic from infective mass lesions. ... 144

xxiv

Table 5. 4: ROC analysis of Type-4 normalized APT-w contrast corresponding to ROI-2 to differentiate low grade glioma (LGG) and infective mass lesions. ... 145