Riemannian geometric statistics in

medical image analysis Pennec X (Ed.)

Visit to download the full and correct content document:
https://ebookmass.com/product/riemannian-geometric-statistics-in-medical-image-ana
lysis-pennec-x-ed/
RIEMANNIAN
GEOMETRIC
S TAT I S T I C S I N
MEDICAL IMAGE
A N A LY S I S
THE ELSEVIER AND MICCAI SOCIETY
BOOK SERIES

Advisory Board
Nicholas Ayache
James S. Duncan
Alex Frangi
Hayit Greenspan
Pierre Jannin
Anne Martel
Xavier Pennec
Terry Peters
Daniel Rueckert
Milan Sonka
Jay Tian
Kevin Zhou

Titles:
Balocco, A., et al., Computing and Visualization for Intravascular Imaging and Computer
Assisted Stenting, 9780128110188
Dalca, A.V., et al., Imaging Genetics, 9780128139684
Depeursinge, A., et al., Biomedical Texture Analysis, 9780128121337
Pennec, X., et al., Riemannian Geometric Statistics in Medical Image Analysis, 9780128147252
Wu, G., and Sabuncu, M., Machine Learning and Medical Imaging, 9780128040768
Zhou K., Medical Image Recognition, Segmentation and Parsing, 9780128025819
Zhou, K., et al., Deep Learning for Medical Image Analysis, 9780128104088
Zhou, K., et al., Handbook of Medical Image Computing and Computer Assisted Intervention,
9780128161760
RIEMANNIAN
GEOMETRIC
S TAT I S T I C S I N
MEDICAL IMAGE
A N A LY S I S

Edited by
XAVIER PENNEC
STEFAN SOMMER
TOM FLETCHER
Academic Press is an imprint of Elsevier
125 London Wall, London EC2Y 5AS, United Kingdom
525 B Street, Suite 1650, San Diego, CA 92101, United States
50 Hampshire Street, 5th Floor, Cambridge, MA 02139, United States
The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, United Kingdom
Copyright © 2020 Elsevier Ltd. All rights reserved.
No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including
photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher.
Details on how to seek permission, further information about the Publisher’s permissions policies and our arrangements with
organizations such as the Copyright Clearance Center and the Copyright Licensing Agency, can be found at our website:
www.elsevier.com/permissions.
This book and the individual contributions contained in it are protected under copyright by the Publisher (other than as may be
noted herein).
Notices
Knowledge and best practice in this field are constantly changing. As new research and experience broaden our understanding,
changes in research methods, professional practices, or medical treatment may become necessary.
Practitioners and researchers must always rely on their own experience and knowledge in evaluating and using any information,
methods, compounds, or experiments described herein. In using such information or methods they should be mindful of their
own safety and the safety of others, including parties for whom they have a professional responsibility.
To the fullest extent of the law, neither the Publisher nor the authors, contributors, or editors, assume any liability for any injury
and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of
any methods, products, instructions, or ideas contained in the material herein.

Library of Congress Cataloging-in-Publication Data

A catalog record for this book is available from the Library of Congress

British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library

ISBN: 978-0-12-814725-2

For information on all Academic Press publications

visit our website at https://www.elsevier.com/books-and-journals

Publisher: Mara Conner

Acquisition Editor: Tim Pitts
Editorial Project Manager: Leticia M. Lima
Production Project Manager: Kamesh Ramajogi
Designer: Miles Hitchen
Typeset by VTeX
 Contents v

Contents
Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvii
Part 1 Foundations of geometric statistics
Chapter 1 Introduction to differential and Riemannian geometry . . . . . . . . 3
Stefan Sommer, Tom Fletcher, Xavier Pennec
1.1 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Riemannian manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.4 Elements of analysis in Riemannian manifolds. . . . . . . . . . . . . . . . . . 20
1.5 Lie groups and homogeneous manifolds . . . . . . . . . . . . . . . . . . . . . . . 24
1.6 Elements of computing on Riemannian manifolds . . . . . . . . . . . . . . 29
1.7 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
1.8 Additional references . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
Chapter 2 Statistics on manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Tom Fletcher
2.1 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.2 The Fréchet mean . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.3 Covariance and principal geodesic analysis. . . . . . . . . . . . . . . . . . . . . 44
2.4 Regression models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.5 Probabilistic models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
Chapter 3 Manifold-valued image processing with SPD matrices . . . . . . 75
Xavier Pennec
3.1 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.2 Exponential, logarithm, and square root of SPD matrices . . . . . . . . 79
3.3 Affine-invariant metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.4 Basic statistical operations on SPD matrices . . . . . . . . . . . . . . . . . . . . 96
3.5 Manifold-valued image processing . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
vi Contents

3.6 Other metrics on SPD matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110

3.7 Applications in diffusion tensor imaging (DTI) . . . . . . . . . . . . . . . . . 117
3.8 Learning brain variability from Sulcal lines . . . . . . . . . . . . . . . . . . . . 122
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
Chapter 4 Riemannian geometry on shapes and diffeomorphisms . . . . . 135
Stephen Marsland, Stefan Sommer
4.1 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
4.2 Shapes and actions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
4.3 The diffeomorphism group in shape analysis . . . . . . . . . . . . . . . . . . 138
4.4 Riemannian metrics on shape spaces . . . . . . . . . . . . . . . . . . . . . . . . . 143
4.5 Shape spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
4.6 Statistics in LDDMM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
4.7 Outer and inner shape metrics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
4.8 Further reading. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
Chapter 5 Beyond Riemannian geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
Xavier Pennec, Marco Lorenzi
5.1 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
5.2 Affine connection spaces. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
5.3 Canonical connections on Lie groups . . . . . . . . . . . . . . . . . . . . . . . . . 182
5.4 Left, right, and biinvariant Riemannian metrics on a Lie group . . 191
5.5 Statistics on Lie groups as symmetric spaces . . . . . . . . . . . . . . . . . . 195
5.6 The stationary velocity fields (SVF) framework for
diffeomorphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
5.7 Parallel transport of SVF deformations . . . . . . . . . . . . . . . . . . . . . . . . 213
5.8 Historical notes and additional references . . . . . . . . . . . . . . . . . . . . . 223
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
Part 2 Statistics on manifolds and shape spaces
Chapter 6 Object shape representation via skeletal models (s-reps)
and statistical analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
 Contents vii

Stephen M. Pizer, Junpyo Hong, Jared Vicory, Zhiyuan Liu, J.S. Marron,
Hyo-young Choi, James Damon, Sungkyu Jung, Beatriz Paniagua,
Jörn Schulz, Ankur Sharma, Liyun Tu, Jiyao Wang
6.1 Introduction to skeletal models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
6.2 Computing an s-rep from an image or object boundary . . . . . . . . 236
6.3 Skeletal interpolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
6.4 Skeletal fitting. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
6.5 Correspondence. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
6.6 Skeletal statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
6.7 How to compare representations and statistical methods . . . . . . . 253
6.8 Results of classification, hypothesis testing, and probability
distribution estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
6.9 The code and its performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
6.10 Weaknesses of the skeletal approach. . . . . . . . . . . . . . . . . . . . . . . . . . 266
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
Chapter 7 Efficient recursive estimation of the Riemannian barycenter
on the hypersphere and the special orthogonal group with
applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
Rudrasis Chakraborty, Baba C. Vemuri
7.1 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
7.2 Riemannian geometry of the hypersphere . . . . . . . . . . . . . . . . . . . . . 274
7.3 Weak consistency of iFME on the sphere . . . . . . . . . . . . . . . . . . . . . . 276
7.4 Experimental results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
7.5 Application to the classification of movement disorders . . . . . . . . 284
7.6 Riemannian geometry of the special orthogonal group . . . . . . . . . 287
7.7 Weak consistency of iFME on so(n). . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
7.8 Experimental results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
7.9 Conclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294
Chapter 8 Statistics on stratified spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
Aasa Feragen, Tom Nye
viii Contents

8.1 Introduction to stratified geometry. . . . . . . . . . . . . . . . . . . . . . . . . . . . 299

8.2 Least squares models. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
8.3 BHV tree space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310
8.4 The space of unlabeled trees . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
8.5 Beyond trees . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 338
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 338
Chapter 9 Bias on estimation in quotient space and correction
methods. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
Nina Miolane, Loic Devilliers, Xavier Pennec
9.1 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
9.2 Shapes and quotient spaces. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344
9.3 Template estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 354
9.4 Asymptotic bias of template estimation . . . . . . . . . . . . . . . . . . . . . . . 359
9.5 Applications to statistics on organ shapes . . . . . . . . . . . . . . . . . . . . . 365
9.6 Bias correction methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 368
9.7 Conclusion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 372
Acknowledgment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
Chapter 10 Probabilistic approaches to geometric statistics. . . . . . . . . . 377
Stefan Sommer
10.1 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377
10.2 Parametric probability distributions on manifolds . . . . . . . . . . . . . . 380
10.3 The Brownian motion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 384
10.4 Fiber bundle geometry. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
10.5 Anisotropic normal distributions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393
10.6 Statistics with bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 398
10.7 Parameter estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 400
10.8 Advanced concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 407
10.9 Conclusion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413
10.10 Further reading. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 414
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 414
 Contents ix

Chapter 11 On shape analysis of functional data . . . . . . . . . . . . . . . . . . . . 417

Ruiyi Zhang, Anuj Srivastava
11.1 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 417
11.2 Registration problem and elastic approach . . . . . . . . . . . . . . . . . . . . 419
11.3 Shape space and geodesic paths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 426
11.4 Statistical summaries and principal modes of shape variability . 430
11.5 Summary and conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 434
Appendix: Mathematical background . . . . . . . . . . . . . . . . . . . . . . . . . 434
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 436
Part 3 Deformations, diffeomorphisms and their applications
Chapter 12 Fidelity metrics between curves and surfaces: currents,
varifolds, and normal cycles. . . . . . . . . . . . . . . . . . . . . . . . . . . . 441
Nicolas Charon, Benjamin Charlier, Joan Glaunès, Pietro Gori,
Pierre Roussillon
12.1 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 441
12.2 General setting and notations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 443
12.3 Currents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 443
12.4 Varifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 450
12.5 Normal cycles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 457
12.6 Computational aspects. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 465
12.7 Conclusion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 472
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 473
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 473
Chapter 13 A discretize–optimize approach for LDDMM registration . . 479
Thomas Polzin, Marc Niethammer, François-Xavier Vialard,
Jan Modersitzki
13.1 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 479
13.2 Background and related work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 483
13.3 Continuous mathematical models . . . . . . . . . . . . . . . . . . . . . . . . . . . . 488
13.4 Discretization of the energies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 493
13.5 Discretization and solution of PDEs . . . . . . . . . . . . . . . . . . . . . . . . . . . 496
13.6 Discretization in multiple dimensions . . . . . . . . . . . . . . . . . . . . . . . . . 506
x Contents

13.7 Multilevel registration and numerical optimization . . . . . . . . . . . . . 512

13.8 Experiments and results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 515
13.9 Discussion and conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 523
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 524
Chapter 14 Spatially adaptive metrics for diffeomorphic image
matching in LDDMM. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 533
Laurent Risser, François-Xavier Vialard
14.1 Introduction to LDDMM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 533
14.2 Sum of kernels and semidirect product of groups . . . . . . . . . . . . . . 539
14.3 Sliding motion constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 545
14.4 Left-invariant metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 548
14.5 Open directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 551
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 554
Chapter 15 Low-dimensional shape analysis in the space of
diffeomorphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 557
Miaomiao Zhang, Polina Golland, William M. Wells III, Tom Fletcher
15.1 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 557
15.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 559
15.3 PPGA of diffeomorphisms. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 563
15.4 Inference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 564
15.5 Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 568
15.6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 569
15.7 Discussion and conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 571
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 574
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 574
Chapter 16 Diffeomorphic density registration . . . . . . . . . . . . . . . . . . . . . . 577
Martin Bauer, Sarang Joshi, Klas Modin
16.1 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 577
16.2 Diffeomorphisms and densities. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 578
16.3 Diffeomorphic density registration . . . . . . . . . . . . . . . . . . . . . . . . . . . . 581
16.4 Density registration in the LDDMM-framework. . . . . . . . . . . . . . . . . 582
16.5 Optimal information transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 585
 Contents xi

16.6 A gradient flow approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 589

Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 600
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 601
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 605
 Contributors xiii

Contributors

Martin Bauer
Florida State University, Department of Mathematics, Tallahassee, FL, United States
Rudrasis Chakraborty
University of Florida, CISE Department, Gainesville, FL, United States
Benjamin Charlier
IMAG, Univ. Montpellier, CNRS, Montpellier, France
Institut du Cerveau et de la Moëlle Épinière, ARAMIS, Paris, France
Nicolas Charon
Johns Hopkins University, Center of Imaging Sciences, Baltimore, MD, United States
Hyo-young Choi
UNC, Chapel Hill, NC, United States
James Damon
UNC, Chapel Hill, NC, United States
Loic Devilliers
Université Côte d’Azur and Inria, Epione team, Sophia Antipolis, France
Aasa Feragen
University of Copenhagen, Department of Computer Science, Copenhagen, Denmark
Tom Fletcher
University of Virginia, Departments of Electrical & Computer Engineering and Computer
Science, Charlottesville, VA, United States
Joan Glaunès
MAP5, Université Paris Descartes, Paris, France
Polina Golland
Massachusetts Institute of Technology, Computer Science and Artificial Intelligence Lab,
Cambridge, MA, United States
Pietro Gori
Télécom ParisTech, LTCI, équipe IMAGES, Paris, France
Junpyo Hong
UNC, Chapel Hill, NC, United States
Sarang Joshi
University of Utah, Department of Bioengineering, Scientific Computing and Imaging Institute,
Salt Lake City, UT, United States
xiv Contributors

Sungkyu Jung
Seoul National University, Seoul, Republic of Korea
Zhiyuan Liu
UNC, Chapel Hill, NC, United States
Marco Lorenzi
Université Côte d’Azur and Inria, Epione team, Sophia Antipolis, France
J.S. Marron
UNC, Chapel Hill, NC, United States
Stephen Marsland
Victoria University of Wellington, School of Mathematics and Statistics, Wellington,
New Zealand
Nina Miolane
Université Côte d’Azur and Inria, Epione team, Sophia Antipolis, France
Stanford University, Department of Statistics, Stanford, CA, United States
Jan Modersitzki
Institute of Mathematics and Image Computing, University of Lübeck, Lübeck, Germany
Fraunhofer MEVIS, Lübeck, Germany
Klas Modin
Chalmers University of Technology and the University of Gothenburg, Department of
Mathematical Sciences, Göteborg, Sweden
Marc Niethammer
Department of Computer Science, University of North Carolina at Chapel Hill, Chapel Hill, NC,
United States
Biomedical Research Imaging Center (BRIC), Chapel Hill, NC, United States
Tom Nye
Newcastle University, School of Mathematics, Statistics and Physics, Newcastle upon Tyne,
United Kingdom
Beatriz Paniagua
UNC, Chapel Hill, NC, United States
Xavier Pennec
Université Côte d’Azur and Inria, Epione team, Sophia Antipolis, France
Stephen M. Pizer
UNC, Chapel Hill, NC, United States
Thomas Polzin
Institute of Mathematics and Image Computing, University of Lübeck, Lübeck, Germany
 Contributors xv

Laurent Risser
Institut de Mathématiques de Toulouse, CNRS, Université de Toulouse, UMR CNRS 5219,
Toulouse, France
Pierre Roussillon
ENS Cachan, CNRS, Université Paris-Saclay, CMLA, Cachan, France
Jörn Schulz
Arctic University of Norway, Tromsø, Norway
Ankur Sharma
UNC, Chapel Hill, NC, United States
Stefan Sommer
University of Copenhagen, Department of Computer Science, Copenhagen, Denmark
Anuj Srivastava
Florida State University, Tallahassee, FL, United States
Liyun Tu
UNC, Chapel Hill, NC, United States
Baba C. Vemuri
University of Florida, CISE Department, Gainesville, FL, United States
François-Xavier Vialard
Laboratoire d’informatique Gaspard Monge, Université Paris-Est Marne-la-Vallée, UMR CNRS
8049, Champs sur Marne, France
Jared Vicory
UNC, Chapel Hill, NC, United States
Jiyao Wang
UNC, Chapel Hill, NC, United States
William M. Wells III
Harvard Medical School, Department of Radiology, Boston, MA, United States
Miaomiao Zhang
Washington University in St. Louis, Computer Science and Engineering, St. Louis, MO,
United States
Ruiyi Zhang
Florida State University, Tallahassee, FL, United States
 Introduction xvii

Introduction

Introduction
Over the last two decades, there has been a growing need in
the medical image computing community for principled meth-
ods to process nonlinear geometric data. Typical examples of data
in this domain include organ shapes and deformations resulting
from segmentation and registration in computational anatomy,
and symmetric positive definite matrices in diffusion imaging.
In this context, Riemannian geometry has gradually been estab-
lished as one the most powerful mathematical and computational
paradigms.
This book aims at being an introduction to and a reference
on Riemannian geometric statistics and its use in medical image
analysis for researchers and graduate students. The book provides
both descriptions of the core methodology and presentations of
state-of-the-art methods used in the field. We wish to present this
combination of foundational material and current research to-
gether with examples, applications, and algorithms in a volume
that is edited and authored by the leading researchers in the field.
In addition, we wish to provide an overview of current research
challenges and future applications.
Beyond medical image computing, the methods described in
this book may also apply to other domains such as signal process-
ing, computer vision, geometric deep learning, and other domains
where statistics on geometric features appear. As such, the pre-
sented core methodology takes its place in the field of geometric
statistics, the statistical analysis of data being elements of nonlin-
ear geometric spaces. We hope that both the foundational mate-
rial and the advanced techniques presented in the later parts of
the book can be useful in domains outside medical imaging and
present important applications of geometric statistics methodol-
ogy.

Contents
Part 1 of this edited volume describes the foundations of Rie-
mannian geometric computing methods for statistics on mani-
folds. The book here emphasizes concepts rather than proofs with
the goal of providing graduate students in computer science the
xviii Introduction

mathematical background needed to start in this domain. Chap-

ter 1 presents an introduction to differential, Riemannian and
Lie group geometry, and chapter 2 covers statistics on manifolds.
Chapters 3–5 present introductions to geometry of SPD matrices,
shape analysis through the action of the diffeomorphism group,
and geometry and statistical analysis beyond the Riemannian set-
ting when an affine connection, not a metric, is available.
Part 2 includes contributions from leading researchers in the
field on applications of statistics on manifolds and shape spaces
in medical image computing. In chapter 6, Stephen Pizer, Steve
Marron, and coauthors describe shape representation via skele-
tal models and how this allows application of nonlinear statisti-
cal methods on shape spaces. Chapter 7 by Rudrasis Chakraborty
and Baba Vemuri concerns estimation of the iterative Riemannian
barycenter, a candidate for the generalization of the Euclidean
mean value on selected manifolds. In chapter 8, Aasa Feragen and
Tom Nye discuss the statistics on stratified spaces that generalize
manifold by allowing variation of the topological structure. Esti-
mation of templates in quotient spaces is the topic of chapter 9
by Nina Miolane, Loic Devilliers, and Xavier Pennec. Stefan Som-
mer discusses parametric statistics on manifolds using stochastic
processes in chapter 10. In chapter 11, Ruiyi Zhang and Anuj Sri-
vastava consider shape analysis of functional data using elastic
metrics.
Part 3 of the book focuses on diffeomorphic deformations and
their applications in shape analysis. Nicolas Charon, Benjamin
Charlier, Joan Glaunès, Pierre Roussillon, and Pietro Gori present
currents, varifolds, and normal cycles for shape comparison in
chapter 12. Numerical aspects of large deformation registration
is discussed in chapter 13 by Thomas Polzin, Marc Niethammer,
François-Xavier Vialard, and Jan Modersitzki. Francois-Xavier and
Laurent Risser present spatially varying metrics for large defor-
mation matching in chapter 14. Chapter 15 by Miaomiao Zhang,
Polina Golland, William M. Wells, and Tom Fletcher presents a
framework for low-dimensional representations of large deforma-
tions and its use in shape analysis. Finally, in chapter 16, Martin
Bauer, Sarang Joshi, and Klas Modin study densities matching in
the diffeomorphic setting.

We are extremely grateful for this broad set of excellent contribu-

tions to the book by leading researchers in the field, and we hope
that the book in its entirety will inspire new developments and
research directions in this exciting intersection between applied
mathematics and computer science.
 Introduction xix

The editors:
Xavier Pennec
University Côte d’Azur and Inria, Sophia Antipolis, France
Stefan Sommer
DIKU, University of Copenhagen, Copenhagen, Denmark
Tom Fletcher
University of Virginia, Charlottesville, VA, United States
February, 2019
Introduction to differential and
1
Riemannian geometry
Stefan Sommera , Tom Fletcherb , Xavier Pennecc
a University of Copenhagen, Department of Computer Science, Copenhagen,

Denmark. b University of Virginia, Departments of Electrical & Computer

Engineering and Computer Science, Charlottesville, VA, United States.
c Université Côte d’Azur and Inria, Epione team, Sophia Antipolis, France

1.1 Introduction
When data exhibit nonlinearity, the mathematical description
of the data space must often depart from the convenient linear
structure of Euclidean vector spaces. Nonlinearity prevents global
vector space structure, but we can nevertheless ask which mathe-
matical properties from the Euclidean case can be kept while still
preserving the accurate modeling of the data. It turns out that in
many cases, local resemblance to a Euclidean vector space is one
such property. In other words, up to some approximation, the data
space can be linearized in limited regions while forcing a linear
model on the entire space would introduce too much distortion.
The concept of local similarity to Euclidean spaces brings us
exactly to the setting of manifolds. Topological, differential, and
Riemannian manifolds are characterized by the existence of local
maps, charts, between the manifold and a Euclidean space. These
charts are structure preserving: They are homeomorphisms in the
case of topological manifolds, diffeomorphisms in the case of dif-
ferential manifolds, and, in the case of Riemannian manifolds,
they carry local inner products that encode the non-Euclidean ge-
ometry.
The following sections describe these foundational concepts
and how they lead to notions commonly associated with geome-
try: curves, length, distances, geodesics, curvature, parallel trans-
port, and volume form. In addition to the differential and Rieman-
nian structure, we describe one extra layer of structure, Lie groups
that are manifolds equipped with smooth group structure. Lie
groups and their quotients are examples of homogeneous spaces.
The group structure provides relations between distant points on
the group and thereby additional ways of constructing Rieman-
nian metrics and deriving geodesic equations.
Riemannian Geometric Statistics in Medical Image Analysis 3
https://doi.org/10.1016/B978-0-12-814725-2.00008-X
Copyright © 2020 Elsevier Ltd. All rights reserved.
4 Chapter 1 Introduction to differential and Riemannian geometry

Topological, differential, and Riemannian manifolds are of-

ten covered by separate graduate courses in mathematics. In this
much briefer overview, we describe the general concepts, often
sacrificing mathematical rigor to instead provide intuitive reasons
for the mathematical definitions. For a more in-depth introduc-
tion to geometry, the interested reader may, for example, refer
to the sequence of books by John M. Lee on topological, differ-
entiable, and Riemannian manifolds [17,18,16] or to the book on
Riemannian geometry by do Carmo [4]. More advanced references
include [15], [11], and [24].

1.2 Manifolds
A manifold is a collection of points that locally, but not glob-
ally, resembles Euclidean space. When the Euclidean space is of
finite dimension, we can without loss of generality relate it to Rd
for some d > 0. The abstract mathematical definition of a man-
ifold specifies the topological, differential, and geometric struc-
ture by using charts, maps between parts of the manifold and Rd ,
and collections of charts denoted atlases. We will discuss this con-
struction shortly, however, we first focus on the case where the
manifold is a subset of a larger Euclidean space. This viewpoint is
often less abstract and closer to our natural intuition of a surface
embedded in our surrounding 3D Euclidean space.
Let us exemplify this by the surface of the earth embedded in
R3 . We are constrained by gravity to live on the surface of the earth.
This surface seems locally flat with two dimensions only, and we
use two-dimensional maps to navigate the surface. When travel-
ing far, we sometimes need to change from one map to another.
We then find charts that overlap in small parts, and we assume
that the charts provide roughly the same view of the surface in
those overlapping parts. For a long time, the earth was even con-
sidered to be flat because its curvature was not noticeable at the
scale at which it was observed. When considering the earth sur-
face as a two-dimensional restriction of the 3D ambient space, the
surface is an embedded submanifold of R3 . On the other hand,
when using maps and piecing the global surface together using
the compatibility of the overlapping parts, we take the abstract
view using charts and atlases.

1.2.1 Embedded submanifolds

Arguably the simplest example of a two-dimensional manifold
is the sphere S2 . Relating to the previous example, when embed-
ded in R3 , we can view it as an idealized model for the surface of
 Chapter 1 Introduction to differential and Riemannian geometry 5

the earth. The sphere with radius 1 can be described as the set of
unit vectors in R3 , that is, the set

S2 = {(x 1 , x 2 , x 3 ) ∈ R3 | (x 1 )2 + (x 2 )2 + (x 3 )2 = 1} . (1.1)

Notice from the definition of the set that all points of S2 satisfy the
equation (x 1 )2 + (x 2 )2 + (x 3 )2 − 1 = 0. We can generalize this way
of constructing a manifold to the following definition.

Definition 1.1 (Embedded manifold). Let F : Rk → Rm be a differ-

entiable map such that the Jacobian matrix dF (x) = ( ∂x∂ j F i (x))ij
has constant rank k − d for all x ∈ F −1 (0). Then the zero-level set
M = F −1 (0) is an embedded manifold of dimension d.

Figure 1.1. An embedded manifold arises as the zero-level subset M = F −1 (0) of

the map F : Rk → Rm . Here F : R3 → R is given by the sphere equation
x → (x 1 )2 + (x 2 )2 + (x 3 )2 − 1, and the manifold M = S2 is of dimension
3 − 1 = 2.

The map F is said to give an implicit representation of the

manifold. In the previous example, we used the definition with
F (x) = (x 1 )2 + (x 2 )2 + (x 3 )2 − 1 (see Fig. 1.1).
The fact that M = F −1 (0) is a manifold is often taken as the
consequence of the submersion level set theorem instead of a
definition. The theorem states that with the above assumptions,
M has a manifold structure as constructed with charts and atlases.
In addition, the topological and differentiable structure of M is in
a certain way compatible with that of Rk letting us denote M as
embedded in Rk . For now, we will be somewhat relaxed about the
details and use the construction as a working definition of what
we think of as a manifold.
The map F can be seen as a set of m constraints that points
in M must satisfy. The Jacobian matrix dF (x) at a point in x ∈ M
linearizes the constraints around x, and its rank k − d indicates
6 Chapter 1 Introduction to differential and Riemannian geometry

how many of them are linearly independent. In addition to the

unit length constraints of vectors in R3 defining S2 , additional ex-
amples of commonly occurring manifolds that we will see in this
book arise directly from embedded manifolds or as quotients of
embedded manifolds.
Example 1.1. d-dimensional spheres Sd embedded in Rd+1 . Here
we express the unit length equation generalizing (1.1) by

Sd = {x ∈ Rn+1 | x2 − 1 = 0} . (1.2)

The squared norm x2 is the standard squared Euclidean norm

on Rd+1 .
Example 1.2. Orthogonal  matrices
 O(k) on Rk have the property
that the inner products Ui , Uj of columns Ui , Uj of the matrix
U ∈ M(k,k) vanish for i = j and equal 1 for i = j . This gives k 2 con-
straints, and O(k) is thus an embedded manifold in M(k,k) by the
equation
  
O(k) = U ∈ M(k,k)  U U  − Idk = 0 (1.3)

with Idk being the identity matrix on Rk . We will see in Sec-

tion 1.7.3 that the rank of the map F (U ) = U U  − Idk is k(k+1)
2 on
k(k−1)
O(k), and it follows that O(k) has dimension 2 .

1.2.2 Charts and local euclideaness

We now describe how charts, local parameterizations of the
manifold, can be constructed from the implicit representation
above. We will use this to give a more abstract definition of a dif-
ferentiable manifold.
When navigating the surface of the earth, we seldom use
curved representations of the surface but instead rely on charts
that give a flat, 2D representation of regions limited in extent. It
turns out that this analogy can be extended to embed manifolds
with a rigorous mathematical formulation.
Definition 1.2. A chart on a d-dimensional manifold M is a dif-
feomorphic mapping φ : U → Ũ from an open set U ⊂ M to an
open set Ũ ⊆ Rd .
The definition exactly captures the informal idea of represent-
ing a local part of the surface, the open set U , with a mapping to a
Euclidean space, in the surface case R2 (see Fig. 1.2).
When using charts, we often say that we work in coordi-
nates. Instead of accessing points on M directly, we take a chart
 Chapter 1 Introduction to differential and Riemannian geometry 7

Figure 1.2. Charts φ : U → Ũ and ψ : V → Ṽ , members of the atlas covering the

manifold M, from the open sets U, V ⊂ M to open sets Ũ , Ṽ of Rd , respectively.
The compatibility condition ensures that φ and ψ agree on the overlap U ∩ V
between U and V in the sense that the composition ψ ◦ φ −1 is a differentiable
map.

φ : U → Ũ and use points in φ(U ) ⊆ Rd instead. This gives us the

convenience of having a coordinate system present. However, we
need to be aware that the choice of the coordinate system affects
the analysis, both theoretically and computationally. When we say
that we work in coordinates x = (x 1 , . . . , x d ), we implicitly imply
that there is a chart φ such that φ −1 (x) is a point on M.
It is a consequence of the implicit function theorem that em-
bedded manifolds have charts. Proving it takes some work, but we
can sketch the idea in the case of the implicit representation map
F : Rk → Rm having Jacobian with full rank m. Recall the setting
of the implicit function theorem (see e.g. [18]): Let F : Rd+m → Rm
be continuously differentiable and write (x, y) ∈ Rd+m such that x
denotes the first d coordinates and y the last m coordinates. Let
dy F denote the last m columns of the Jacobian matrix dF , that is,
the derivatives of F taken with respect to variations in y. If dy F
has full rank m at a point (x, y) where F (x, y) = 0, then there ex-
ists an open neighborhood Ũ ⊆ Rd of x and a differentiable map
g : Ũ → Rm such that F (x, g(x)) = 0 for all x ∈ Ũ .
The only obstruction to using the implicit function theorem di-
rectly to find charts is that we may need to rotate the coordinates
on Rd+m to find coordinates (x, y) and a submatrix dy F of full rank.
With this in mind, the map g ensures that F (x, g(x)) = 0 for all
x ∈ Ũ , that is, the points (x, g(x)), x ∈ Ũ are included in M. Set-
ting U = g(Ũ ), we get a chart φ : U → Ũ directly by the mapping
(x, g(x)) → x.
8 Chapter 1 Introduction to differential and Riemannian geometry

1.2.3 Abstract manifolds and atlases

We now use the concept of charts to define atlases as collec-
tions of charts and from this the abstract notion of a manifold.
Definition 1.3 (Atlas). An atlas of a set M is a family of charts
(φi )i=1,...,N , φi : Ui → Ũi such that
• φi cover M: For each x ∈ M, there exists i ∈ {1, . . . , N } such that
x ∈ Ui ,
• φi are compatible: For each pair i, j ∈ {1, . . . , N } where Ui ∩ Uj
is nonempty, the composition φi ◦ φj−1 : φj (Ui ∩ Uj ) → Rd is a
differentiable map.
An atlas thus ensures the existence of at least one chart cover-
ing a neighborhood of each point of M. This allows the topological
and differential structure of M to be given by a definition from the
topology and differential structure of the image of the charts, that
is, Rd . Intuitively, the structure coming from the Euclidean spaces
Rd is pulled back using φi to the manifold. In order for this con-
struction to work, we must ensure that there is no ambiguity in
the structure we get if the domain of multiple charts cover a given
point. The compatibility condition ensures exactly that.
Definition 1.4 (Manifold). Let M be a set with an atlas (φi )i=1,...,N
with φi : Ui → Ũi , Ũi ⊆ Rd . Then M is a manifold of dimension d.
Remark 1.1. Until now, we have been somewhat loose in describ-
ing maps as being “differentiable”. The differentiability of maps on
a manifold comes from the differential structure, which in turn is
defined from the atlas and the charts mapping to Rd . The differ-
ential structure on Rd allows derivatives up to any order, but the
charts may not support this when transferring the structure to M.
To be more precise, in the compatibility condition, we require the
compositions φi ◦ φj−1 to be C r as maps from Rd to Rd for some
integer r. This gives a differentiable structure on M of the same
order. In particular, when r ≥ 1, we say that M is a differentiable
manifold, and M is smooth if r = ∞. We may also require only
r = 0, in which case M is a topological manifold with no differ-
entiable structure.
Because of the implicit function theorem, embedded submani-
folds in the sense of Definition 1.1 have charts and atlases. Embed-
ded submanifolds are therefore particular examples of abstract
manifolds. In fact, this goes both ways: The Whitney embedding
theorem states that any d-dimensional manifold can be embed-
ded in Rk with k ≤ 2d so that the topology is induced by the one of
the embedding space. For Riemannian manifolds defined later on,
this theorem only provides a local C 1 embedding and not a global
smooth embedding.
 Chapter 1 Introduction to differential and Riemannian geometry 9

Example 1.3. The projective space Pd is the set of lines through

the origin in Rd+1 . Each such line intersects the sphere Sd in two
points that are antipodal. By identifying such points, expressed by
taking the quotient using the equivalence relation x ∼ −x, we get
the representation Pd  Sd / ∼. Depending on the properties of the
equivalence relation, the quotient space of a manifold may not be
a manifold in general (more details will be given in Chapter 9).
In the case of the projective space, we can verify the above ab-
stract manifold definition. Therefore the projective space cannot
be seen as an embedded manifold directly, but it can be seen as
the quotient space of an embedded manifold.

1.2.4 Tangent vectors and tangent space

As the name implies, derivatives lies at the core of differential
geometry. The differentiable structure allows taking derivatives of
curves in much the same way as the usual derivatives in Euclidean
space. However, spaces of tangent vectors to curves behave some-
what differently on manifolds due to the lack of the global refer-
ence frame that the Euclidean space coordinate system gives. We
here discuss derivatives of curves, tangent vectors, and tangent
spaces.
Let γ : [0, T ] → Rk be a differentiable curve in Rk parameter-
ized on the interval [0, T ]. For each t, the curve derivative is
⎛ ⎞
d 1
dt γ (t)
d ⎜ .. ⎟
γ (t) = γ̇ = ⎜
⎝ . ⎠.
⎟ (1.4)
dt
d k
dt γ (t)

This tangent or velocity vector can be regarded as a vector in Rk ,

denoted the tangent vector to γ at t. If M is an embedded mani-
fold in Rk and γ (t) ∈ M for all t ∈ [0, T ], we can regard γ as a curve
in M. As illustrated on Fig. 1.3, the tangent vectors of γ are also
tangential to M itself. The set of tangent vectors to all curves at
x = γ (t) span a d-dimensional affine subspace of Rk that approx-
imates M to the first order at x. This affine space has an explicit
realization as x + ker dF (x) where x = γ (t) is the foot-point and
ker dF denotes the kernel (null-space) of the Jacobian matrix of F .
The space is called the tangent space Tx M of M at the point x. In
the embedded manifold case, tangent vectors thus arise from the
standard curve derivative, and tangent spaces are affine subspaces
of Rk .
On abstract manifolds, the definition of tangent vectors be-
comes somewhat more intricate. Let γ be a curve in the abstract
Another random document with
no related content on Scribd:
 CARLOMAN.

Courage, sire,
Is constant industry for happiness.
When I become a monk——

CHILPERIC.

Nay, no confession,
No putting reasons to your Overlord.
[to his nobles]
You need not shake your spears so stormily,
We leave you a stout leader for your wars,
[to Carloman] And you, your liberty. What use of it
You make is of no moment to the world,
And does not raise my curiosity,
Who for myself have found in meat and drink,
In sleep and long, long abstinence from care
The pleasure proper to me. Pepin, come!

[Exeunt Chilperic, Pepin and the Frankish Nobles.]

CARLOMAN.
He has no sight of God, is imbecile
And dropping into clay. I should not let
This show dishearten me; but I have suffered
A vulgar tongue to tell what from my lips
Alone is truth—that as the hidden spring,
Restless at touch of the diviner’s rod
Is dragged through to the surface by his spells,
I am discovered and borne upward, made
The answer to some perilous appeal:
And for my folly I must be dismissed
By a mere dotard with a passing sigh
Of envy, who forego the battlefield,
The Council-chamber, the sweet clang of arms
For just a pricking wonder at my heart,
A knowledge I would give to secrecy
Plunging it headlong in the ear of God.
Oh for the cloister! I will make escape
At once, in silence, without taking leave:
My joy is in the consciousness that Time
Will never draw me back to any wish
To any fondness I am flinging off....

[Enter Geneviva.]

My wife!
Is Geneviva come to me?

GENEVIVA.
Now the dull monk has left you. Rouse your head!
I have been taking thought how best to trim
My beauty for you. Boniface was slow
In giving counsel; slowly I took up,
Handled and dropt my jewels. Of a sudden,
When Pepin’s voice was heard upon the stair,
I laid these blossoms in a ruddy knot
Thus hasty on my bosom. Come to me.
My lord, you owe me many hours of love,
So many hours I have been beautiful
In vain. You do not see me when I sing,
You miss the marks of music in my face,
You do not love the hunt, and you have never
Ridden beside me in the morning light.
You see me but as now when I am vexed
And haughty for caresses.

CARLOMAN.

[after a pause] Geneviva,

You are a Christian?

GENEVIVA.

Dear my lord, you speak

As if I were laid sick.

CARLOMAN.

You were baptised?

GENEVIVA.
 Assuredly, but the cold font has left
No chill upon my heart. Think not of that,
Think of our marriage-day. You leave me lonely
While Boniface enthralls you.

CARLOMAN.

[with hesitation] Women even

Have put aside their pomps and vanities ...

GENEVIVA.

Oh, leave me, you are insupportable!

You bring me word of kingdoms and of monks,
And thoughts of things that have not come to pass,
Or should be quite forgotten. We could spend
So sweet a moment now, for you are loved,
My Carloman—What need is there of talk
Concerning other matters?—loved of me,
Dreamed of when I am dreaming, when I wake
Wept for, sighed after. I have never cared
To listen to the minstrels, for the praise
My beauty covets most is in your eyes.
How wild they look and solemn!

[Carloman folds her in his arms quietly. Then with great effort bends over
her and speaks]
CARLOMAN.

Marcomir
Is restless for a pilgrimage to Rome.
I think we shall be starting presently:
And afterward ... If I am long away ...

GENEVIVA.
 [breaking from him]
Oh, think a little! Can you leave this hair
So crisp and burnished? When the sun is bright
Across your shield, it has no livelier flash—
Confess, it has not? But you come to me
Stale, weary from your dreams and abstinence,
And tingle my suspicion.

CARLOMAN.

If these dreams
Were growing all the world to me!—You start,
You turn away, you will not understand.
The fear of hurting you has made me keep
So distant from you lately, and my eyes
You thought were worn with vigil and with books
Have burnt with tears at night for many a month
To think you have not known the tyrant-joy
That moves a soul to change and severance,
Except upon the day when for my sake
You parted from your home: but by the rapture
That made such tumult in the daughter’s grief
When she became a bride, your husband now
Implores your comprehension.
All thou hast,
So the Church teaches, family and spouse,
The child thou hast begotten, thine own life
Thou must abhor, if thou would’st have new days
Of blessing on the Earth. I feel this law
Is written in my very heart of hearts,
There is such haunting freshness deep below
The sorrow of farewell.

GENEVIVA.
[defiantly] My God is Love—
The God who made a bower in Paradise,
Who wedded Eve and Adam, who abode
In the sweet incense of His Church to bless
My marriage.
[Carloman stretches out his hand to support her.]
Have no fear that I shall fall,
I cannot swoon while I remember it—
How in the songful hush a restless hand
Grew tight about my fingers, and a vow
Thrilled all the girl in me to womanhood,
And stung the future lying at my heart
To joy and frankness. That was years ago ...
[She breaks into a bitter laugh]
O Carloman, you know not what you do,
You know not what I am, nor what a blank
Of mercy there is in you!

CARLOMAN.

Were I dead,
You would not be so violent: in a trance
Of resignation you would think of me,
With tears, not gasping laughter.

GENEVIVA.
[pacing the room excitedly] Pilgrimage!
Did you say, pilgrimage? To think of you
Growing each day more cramped about the mouth,
More full of resolution in the eyes.
What shall I do? Pray for you—but the dead,
You have just told me, should be left unmourned,
Forgotten as last summer’s autumn-leaves.
[facing him coldly] My lord, I am no reliquary-urn;
There is no widow in me.

[with still greater change of manner] If you leave

Your Kingdom, there are certain things to do
Before you start. There is that Gothic King,
The captive Hermann—you must break his chains.

CARLOMAN.

Hermann is dead. Count Marcomir reports

Last night he found him lifeless.

GENEVIVA.

[gasping] Late last night?

Marcomir!—Take your fingers from my sleeve;
But summon Marcomir, and if again
There is intelligence to break to me
Likely to hurt, give him the charge of it.

CARLOMAN.
 No, Geneviva. I have little speech;
But when the secret crept into my soul
I loved you, it was not to Marcomir
I spoke: and if another secret now
Is breaking through my nature, do not think
That he will be the spokesman.
[noticing her agitation] Hermann died
I think by his own hand; he courted death.
What can a man prize in captivity?
[as Geneviva grows more agitated]
There! I will speak no more of him. Your maids—
[turning to summon her attendants].

GENEVIVA.

Weave the great arras. They have no concern

With me, except in silence to array.
You thought I cared to gossip with my maids!
But summon Marcomir.

[She looks after Carloman, who walks out, stroking his chin].

To think he dared
To lean above me with those burning eyes
Unconscious what they glassed. I did not learn
From him the magic that was born in me,
I learnt it when great Hermann passed in chains,
And he is dead. I promised I would go
To-day and visit him. How could he die?
[Marcomir enters.]
Why, you are deadly pale!
[She recoils, and says in a faint voice]
It is the hour
Fixed for our visit.
 MARCOMIR.

But the man is dead.

GENEVIVA.

What does he look like now? Is he so changed

I must not see him?

MARCOMIR.

Death is not a fact

To touch with simile. What looks he like?
All men in moonlight mind one of the moon,
All dead men look like death.

GENEVIVA.

He lies in chains?
Are the brows restful?

MARCOMIR.

Had you been a man

You would have asked me how he came to die,
No more!

GENEVIVA.

I had forgotten ... then he perished

As Carloman reports?
[Marcomir turns away.] You cannot bear
That I should mourn him?

MARCOMIR.
 [facing her again] Oh, a lifetime, if
It please you! I am going to a place
Where love is held of little consequence.

GENEVIVA.

Then you are bound for hell.

MARCOMIR.

[between his teeth] But you are safe!

GENEVIVA.

Keep me recluse from love, as men from war,

You spoil my faculties. Where will you go?

MARCOMIR.
To any coast you have not trod, wherever
The flowers are different from the flowers you wear,
To some Italian convent. Geneviva,
I am not framed to see you minister
To other men; but when long years are passed,
It may be in a fresco, I shall find
Some figure of a lady breaking bread
To mendicants, and kneel and pray to her
That she may bless me also: but till then ...
[covering his eyes]
O God, you shall not tempt me, though I feel
Just how your hair burns in a fiery wreath
Above your brow, and how your eyes are soft
With blue, and deeper blue, as through the hills
The valley stretches azure to the close.
You shall not tempt me, though I almost hear
Your bosom taking record of your breath,
And I could sit and watch that tide of life
Rising and falling through the lovely curves,
Till I was lost in ecstasy.

GENEVIVA.

Oh, hush!
But then you love me. It was in a fit ...?

MARCOMIR.

Of devilish malice.

GENEVIVA.

In a jealous fit?
You shall remain.
[She goes up to him: he takes her hands in his, kisses them coldly, and puts
them away.]
MARCOMIR.

I did not answer you—

His face was drawn.

GENEVIVA.

And I had given you charge

Of the great restive soldier.

MARCOMIR.

True, I swerved;
I have confessed my sin, and now must bear
The settling of my spirit on the Cross.

GENEVIVA.

So many favours!

MARCOMIR.

But you kissed his brows—

What need was there of that?

GENEVIVA.

You love me then,

You love me! Would you murder him again
If I again should touch him with my breath?

MARCOMIR.
 Again, again.

GENEVIVA.

And Carloman complains

I am indifferent to him!

MARCOMIR.

He forgets;
But, Geneviva, if a thousand years
Broke over me, when Time had cleared his storms
I should look up and know your face by heart.

GENEVIVA.

Then stay, stay, stay with me!

Have you once thought
Through the long years how it will fare with me—
Nothing to watch except the sullen waste
Of my own beauty? Marcomir, I hold
If there be judgment it shall be required
Of women what delight their golden hair
Has yielded—have they put its wealth to use,
Or suffered it to lie by unenjoyed?
I rather would die spendthrift, nothing left
Of my rich heritage, save memory
Of the wild, passing pleasure it conferred
Than keep it untransmuted. And you choose
To take from me the only eyes that care
To mirror mine! I have so often thought
That some day I shall drown myself: the water
Reflects me with desire.

MARCOMIR.
 [bitterly, as he turns away] A soul so wide
In innocence, so regal, on the day
He wedded, he appointed me your squire!

GENEVIVA.

[following him]
He keeps you with him, you can read his heart,
You know what way he travels, when his soul
Flies homeward. Tell me—’tis the only knowledge
I crave for in the world—does Carloman
Still hold me in affection? I beseech,
Tell me the truth. He loves you——

MARCOMIR.

Yes, he loves,
He does not use me for his purposes.
[perceiving PEPIN]
Not Carloman—his brother on the stair
Laughs at your light behaviour. So you lose
One last poor opportunity.

[Re-enter Pepin.]
PEPIN.

Good even.
Well, my fair sister, you have heard the news,
Wept [glancing at Marcomir]
and found consolation.
But to think
The son of Charles Martel should be a monk!

GENEVIVA.
 A monk!—a pilgrim?

PEPIN.

No, a cloistered monk.

MARCOMIR.

What is his crime?

PEPIN.

Oh, no impiety;
A crazy fit: he must get near to God,
So puts away all intercourse with man:
And while I rule he thinks to thrill the world
With some convulsive movement from his prayers.
Ha, ha! But you shall queen it as before.

GENEVIVA.

Go fetch my husband and remain without,

For he alone can speak to me of this.

[Exit Pepin.]

[turning to Marcomir]
You are a murderer: this act of yours
Will leave me very lonely.

MARCOMIR.
 I repent.

GENEVIVA.

There is no sin like that of looking back

When one has sinned. Whatever one attempts
It perfected in patience brings reward.
My Carloman will prosper: his whole heart
Is gone away from me.
Why there he is,
Passing in zealous talk with Boniface.

[Carloman and Boniface cross from right to left at the back of the hall.
Geneviva intercepts them.]

Farewell!

CARLOMAN.

[arrested] O Geneviva!

GENEVIVA.

Not my name,
Never my name again. Say, holy father—
They take new titles who renounce the world?

CARLOMAN.
 [with flushing eagerness]
Then you too will renounce it? oh, the joy!
There is a strange new passion in your eyes.
Speak to me ... but you cannot! I could take
No leave of you in your fierce, worldly mood;
Now all is changed.

GENEVIVA.

Yes, all. How long ago

It seems since we were married!

CARLOMAN.

Think the day

Is yet to come, the joy is all before.
[taking her face between his hands]
O Boniface, this is no temptress’ face!
God has been with her, and she starts as I
Free in the great endeavour.

BONIFACE.

Do you choose,
Lady, a mere retreat among the nuns,
Or, like your husband, do you break all ties
That bind you to the earth?

GENEVIVA.

They all are broken:

Except ... oh, I forgot! I have a son.

CARLOMAN.
 [nervously]
Pepin will guard him.

GENEVIVA.

Are you dreaming still?

Fool, fool! I tell you Pepin shall decide
What robes I wear, and haply suffer me
Sometimes at entertainments to look on,
And see young Charlemagne praised. But for my child
He shall remain with me.
[Re-enter Pepin] All is confirmed.
I shall not quit the world. How easily
A man is duped with God upon the brain!
I shall continue in my womanhood,
Giving, receiving pleasure.
I have heard
So much and suddenly; for Marcomir
Is to become a monk.
[to Carloman] Give him no welcome.
He takes the cowl a penitent; he is not,
Like you, a white-souled wayfarer.
[to Pepin] How strange
That we must pair together, you and I;
I know so little of your tastes and now
I must be often in your company.

MARCOMIR.

My lord, speak to her.

PEPIN.
 Come, an end to this!
Brother, if you are wise you will not leave
This woman in the world. Convents are made
To tame the pride of such and keep them cool.

CARLOMAN.

O Geneviva, for my sake, and yet....

Not so, beloved.

[He turns away and covers his face.]

GENEVIVA.

Marcomir, farewell!
You will be monks together. When my husband
Forgets me, you must bring me to his thoughts
Recall that day we hunted and you fell;
I stayed to tend you; but the whole live day
My voice rang through the woods for Carloman
Until I wearied you; he was not found;
But you remember how I cried for him.

MARCOMIR.

Consul, have pity on her. I am free,

But she has need of love.

GENEVIVA.
 O insolence!—
The virginal chill heart!—No intercession!
[to Carloman]
Our marriage is dissolved. How great a stranger
You have become to me! I should grow mad
To breathe by you another single hour.
[to Boniface]
And you, old man, who stand with such meek eyes,
Though you have robbed me of my name of wife,
And made my boy an orphan—go your way!
I cannot curse you, but I prophesy:
Dishonour motherhood, plant virgin homes,
Give to religion the sole charge of love,
And you will rear up lust of such an ice
As Death himself will shiver at.
[to Pepin] Lead on!
Now there is hope you may become a King,
There should be some high festival to keep
To-night in everlasting memory.
Lead me away.

PEPIN.

Brother, in all—good luck!

And may the Convent’s fare be angels’ food.
Your wife’s tears soon will dry.

[Exeunt Pepin and Geneviva.]

CARLOMAN.

The thing to do
Is simply just the sole thing to be done.
There should have been no tears, no taking leave;
A freeman can do anything he will.