Tutorials

Tutorial title: Deformable Models in Medical Image Processing;
Advances in Image Guided Radiotherapy

Lecturers: Bogdan J. Matuszewski, Christopher J. Moore
Affiliation:
Bogdan J. Matuszewski - Applied Digital Signal and Image Processing Research Centre Department of Technology, University of Central Lancashire, United Kingdom,
Christopher J. Moore - Christie Hospital NHS Trust, North Western Medical Physics , United Kingdom

Tutorial outline:
The tutorial will describe the state-of-the-art in modelling deformations with applications in medical image processing. The focus of the tutorial will be on the use of deformable techniques in medical image registration, but other areas including segmentation will be also discussed. Image registration and segmentation are key enabling technologies in the medical image processing which form a part of almost any medical image processing algorithm. They are essential to provide common reference frame, to acquire measurements enabling robot assisted interventions, post-surgical assessment, monitoring of disease progress, detection of abnormalities, measurement of radiotherapy radiation dose, or the functional imaging and its mapping to the anatomical structures. Although different applications areas of deformable models in the medical image processing will be examined the main emphasis will be on their use in radiation therapy. Radiation therapy is designed to cure localised cancer by repeatedly targeting a tumour with small doses of high energy radiation over many days. The treatment plan provides a sophisticated estimate of the radiation dose distribution inside the patient, usually as they appear on a computer tomography (CT) scan before treatment. However, some time later, when treatment is in progress and radiation is being directed at the tumour, the options for monitoring the patient's internal anatomy are very limited. The tutorial will draw from new research endeavour undertaken by the authors on digitally 'seeing and measuring' what is happening to the patient during their radiation therapy. Their research project, named Metrology Guided Radiation Therapy (MEGURATH), is developing technologies and methodologies for live measurement of a patient's position and shape, linking this directly to internal anatomy during treatment using optical surface scanner and on-board imager (Image Guided Radiation Therapy)

The tutorial will introduce taxonomy of the deformable methods. The special attention will be put on explaining the underling methodology of using deformable modelling. This will be reinforced by the structure of the tutorial, where key design choices, namely similarity measures, displacement/shape models, and optimisations procedures, will be identified. The main implementation aspects of deformable models will be clearly specified with discussion of different options available for each step, their performance and examples of their use. Throughout the tutorial the ideas introduced will be visualised using practical cases with real multimodality medical imaging data including radiotherapy treatment planning CT scans (RTPCT), cone beam CT scans (CBCT), dynamic magnetic resonance scans (dMR) and optical body surface scanning (OBSS) data. In contrast to rigid/affine registration, the quantitative assessment of the accuracy of deformable registration is a challenging problem. It has not yet been fully solved. Number of different techniques have been proposed for this including use of human body phantoms and simulation of plausible deformations using biophysical modelling techniques. A short overview of these techniques will be included in the tutorial for completeness. The tutorial will conclude with a few case studies to stress overall methodology and to show a very practical nature of deformable models applied in medical imaging.

The tutorial is designed to give an extensive introduction to deformable models, their use, and to stimulate interest in this field. Some of the methods will be explained in more details to give a better taste of their complexity and to satisfy more accomplished participants. No priori knowledge of the medical imaging techniques will be required and although majority of the examples will be drawn from the medical domain the described techniques have much wider applications.

The structure of the tutorial is as follow:

• Introduction
  • Deformable models versus global rigid/affine models.
  • Registration, segmentation, fusion and other uses of the deformable models in medical image processing.
  • Medical imaging modalities and their properties with respect to registration and segmentation.
• Similarity Measures
• Sum of square differences and correlations measures
• Mutual information, normalised mutual information, and correlation ratio
• Explicit modelling of the image intensity
• Deformable Models
  • Feature versus intensity based methods
  • Free form deformations
  • Constraints imposed on deformable models
  • Elastic, plastic, fluid and diffused models
  • Variational models
  • Stochastic models, stochastic shape models and stochastic displacement models, Bayesian methodology, likelihood function, estimation of the priors.
  • Spatio-temporal models
• Optimisation Procedures
  • Numerical optimisation (simplex, steepest descent, conjugate gradients, LM, …)
  • Direct and indirect methods in solving variational problems
  • Finite element method (FEM) versus finite difference method (FDM)
  • EM, Monte Carlo and sequential Monte Carlo methods
• Validation Methods
• Case Studies
  • Spherical harmonics PCA bladder segmentation.
  • Deformable registration between cone-beam CT and radiotherapy planning CT for treatment planning grey scale calibration
  • Dynamical statistical shape priors for level set based organ segmentation and deformable registration.

Information about the lecturers:



Bogdan Matuszewski received MSc (with honors), and PhD degrees, both in electronic engineering from Wrocław University of Technology (Poland). Currently he is a senior lecturer in the Department of Technology, and a Head of the Robotics and Vision Laboratory at the University of Central Lancashire, United Kingdom. He is a Member of the IEEE and BMVA. Dr Matuszewski has published over 50 research papers in different areas of computer vision and image processing. His recent research interests include use of Bayesian methodology for modelling, classification and tracking; deformable models and their applications to data registration and segmentation; estimation of a human posture; and image based rendering. He collaborated with many industrial partners including BAE Systems, Alenia Aerospazio and EADS. Most recently he has been working with number of universities from the Atlantic Arch region on hyper-spectral imaging project PIMHAI; with the computer vision group at the Heriot-Watt University on multi-view representation for view synthesis; and with Christie Hospital and Liverpool John Moores University on Metrology Guided Radiation Therapy project.




Prof. Christopher J. Moore, North Western Medical Physics, Christie Hospital, Manchester M20 4BX
Gained a 1st class honours degree in Physics from Manchester University in 1976 and joined North Western Medical Physics at the Christie Hospital, where he now leads the Developing Technologies Section of Radiotherapy Physics. He is a Chartered Scientist and State Registered Clinical Scientist. His M.Sc is in Computational Physics, which he obtained from Salford University in 1982. He obtained his Ph.D in Image Analysis from Manchester University in 1988. He holds a visiting chair in Medical Physics at Liverpool John Moores University.
He has been responsible for research in clinical-signal analysis, diagnostic and radiotherapy image processing, image-guided irradiation of tumours, dosimetric and endocrine modelling, and radiobiological prediction. He developed clinical facilities for CT-image assisted planning of cervical cancer therapy and led the creation of image and graphics based computerised conformal radiotherapy using the Western world's first multi-leaf collimator for shaping mega-voltage X-ray treatment beams. Some 10,000 patients were treated with deliverables from these extended programmes.
He has led or participated in nine UK and European research collaborations in the past decade, is an EU Expert scientific evaluator and a reviewer for the Engineering and Physical Science Research Council. He has over 100 peer reviewed publications.