Ebook: Contemporary Perspectives in Three-Dimensional Biomedical Imaging

Images are everywhere in Biology and Medicine. The physician, the surgeon, and the biologist use them on a routine basis. Their importance is recognized for numerous reasons. 3-D images allow better morphological and functional assessment of the anatomy and the physiology of the human body to be performed. 3-D imaging is a powerful tool for diagnosis (detection, characterization, quantification of lesions, estimation of blood flow, functional assessment, etc.) and therapy (3-D localization, volume computation, surgical planning, etc.). A thorough checkup of a patient can now be performed by means of different imaging modalities. Many believe that echography is the stethoscope of tomorrow. Even palpation could be made through an imaging technique, like the recently reported Magnetic Resonance Elastography approach, which can be used to study and visualize strain-wave propagation within objects or material.

Images play a key role in Biology. They are crucial in the analysis of a wide variety of biological materials ranging from tissues to macromolecules. The definitive advantage of imaging is its accuracy and the diversity of vision systems operating at various resolutions up to the atomic level. 3-D reconstructed images in Electron Microscopy, for example, offer the only way of seeing isolated DNA molecules.

Looking backward over the major achievements in the field of Medical Imaging, it is clear that technological advances have been made possible only under specific conditions, when sufficient multidisciplinary competences have been brought together. The invention of the Computed Tomography scanner in the late 60s is a success story of a neuroradiologist (J. Ambrose) and a physicist (G.N. Hounsfield). A physician (R.V. Damadian) was the first to think that Nuclear Magnetic Resonance could be used for medical imaging purposes. But MRI became a reality when P.C. Lauterbur, a physicist, solved the localization problem in the mid-70s using magnetic field gradients.

Today, the main challenges still consist in developing integrated applications in a clinical environment. Designing, implementing and validating complex systems necessitates medical knowledge, in depth involvement of clinical partners, basic and applied technological research. Multidisciplinarity is the key to success. Education and training are necessary too.

This book is intended to provide a convenient reference for the results and current trends in the field of Biomedical Imaging. It brings together in one volume eleven chapters concerning the different aspects of this field in terms of basic engineering issues (reconstruction, visualization, segmentation, texture, modeling, interpretation, and fusion) and biological and medical issues (the study of macromolecules, X-ray imaging, ultrasound imaging, computed tomography, emission computed tomography, surgery, morphology, anthropometrics, forensic science, computer integrated surgery and therapy, and clinical evaluation).

A review of the general approach and major applications is provided in the introductory chapter. The other topics are addressed in dedicated chapters written by leading European researchers in Biomedical Imaging. We believe that they provide the reader with up-to-date, state-of-the-art knowledge on current and emerging areas. The appendix presents an accessible survey of differential geometry.

We are especially grateful to the authors of the different chapters who enthusiastically joined in with the edition of this book and who have provided an excellent and recent analysis and view of their own scientific field. We are also deeply indebted to co-workers, collaborators and doctoral researchers of the Laboratoire de Traitement du Signal et de l' Image, the department of Image et Traitement de l'Information and of the Laboratoire de Traitement de l'Information Médicale, who are responsible for significant contributions to the first chapter.

This book would not have been made without the initiative of Dr E. Fredriksson, Director, IOS Press. The help of Drs Brinkman and Hermsen, assistant publishers, IOS Press is also acknowledged, as well as the contribution of Janet Ormrod, ENST de Bretagne, who carefully proofread part of the manuscript.

Brest and Rennes, Brittany, France. November 1996

Christian Roux and Jean-Louis Coatrieux

Dedicace

To Claire, Benoît, Etienne and François, Christian Roux

To my loved ones, Jean-Louis Coatrieux

This chapter will introduce the issues of major importance in medical imaging. These elements will be examined first through the technological evolution and the clinical areas of concern. After a brief overview of medical imaging issues, the third section is devoted to image analysis going from reconstruction, segmentation, and object extraction up to object modeling (shape description and metrics derivation). These key steps, extensively developed in the other chapters of the book, raise a number of questions about our capability to detect, precisely delineate the organs and characterize the lesions from one imaging modality or multiple sources (i.e. multimodal imaging where image fusion and the accompanying registration problems have to be solved). The next section brings more insights into three topics not covered in the sequel of the book : (i) motion analysis in 2-D or 3-D image sequences; (ii) virtual reality which allows medical operations to be simulated, planned and controled, and (iii) the evolution from intermediate to high level description, a necessary step between low level and interpretation level. They suppose high technology environments and also efficient solutions to realistically view, track and act on the organs.

Fully three-dimensional image reconstuction is a major challenge for new tomographic devices in radiology and nuclear medicine. The use of a 2-D detector such as a radiographic image intensifier or an Anger camera provides series of 2-D projections which are processed by the reconstruction software to compute the unknown 3-D image. In cone-beam geometry, all the projection lines converge to a same focal point. This geometry provides a zoom effect. But dedicated reconstruction algorithms are needed since a given line crosses several volume slices. In this chapter, we give an overview of cone-beam reconstruction algorithms, with a special emphasis on transform methods and in particular on indirect algorithms via the Radon domain. We summarize the mathematical framework, and we provide some experimental results using the RADON software. Next, we consider two specific research topics in cone-beam reconstruction which are the processing of the shadow zone in the Radon domain associated with a circular acquisition trajectory and the reconstruction of a region of interest within the object using either a dual magnification acquisition or a multifocal geometry.

This chapter addresses the issue of three-dimensional reconstruction of biological macromolecules from electron microscopy images. First, the electron microscope device, the micrographs as specimen projections and the projection collection geometry are discussed. After that, a brief introduction to the problems involved in sample preparation and observation is made. In a third section the different three-dimensional reconstruction methods are presented, and finally, the chapter finishes with an example in which a protein is reconstructed starting from micrographs and ending with a nice and very informative volume.

Since the mid seventies many researchers are dealing with the problem of non invasive diagnosis. One way of doing this are the imaging techniques used in almost every clinical environment, e.g. classic X-rays, ultrasound, Computed Tomography (CT), Magnetic Resonance Imaging (MR), or more experimental techniques like PET or SPECT. The large amount of information contained in such images produced with imaging devices of the above categories is not easy to analyse. Hence, the images obtained have to be processed in a way that will yield 3-D-looking images similar to those in anatomical atlases. This chapter aims at the presentation of the most important methods in medical 3-D-visualization and some representative work for each method discussed.

Furthermore there will be a description of some possibilities to compensate for the drawbacks of visualization that include speed, integration in clinical routine and non automatic segmentation. For speeding up the visualization of volume images some parallelization techniques will be presented. In the last part of this chapter the reader will find some guidelines for the design of a user-centered graphical interface in order to allow less experienced computer users, like medical personnel, to use some visualization algorithms in a comprehensive and less time-consuming way. Such interfaces might allow a fast integration in clinical routine.

As the following presentation of 3-D visualization in medicine is supposed to be a survey that will serve as an introduction to the concepts, we avoid presenting mathematical formulas for the models presented and concentrate on a verbal description.

We show how to extract reliable informations about the shape of 3-D objects, obtained from volume medical images. We present an optimal region-growing strategy, that makes use of the differential characteristics of the object surface, and achieves a stable segmentation into a set of patches of quadratic surfaces. We show how this segmentation can be used to recognize and locate a target sub-structure on a global anatomic structure.

The main objective of any diagnostic study of image is the characterization of tissues. Therefore images are acquired in order to determine whether the tissues in the selected area for study show normal (healthy tissue) or pathological characteristics. Hence, giving the exact appearance of the tissue in the image, pathology is classified in the most precise way possible, by refering to other diagnostic data. This process requires both a complex evaluation of the various properties which characterize the appearance of the image studied and the comparison of these properties with the radiologist's experimental database. Giving the subjective nature inherent in the majority of appreciations associated with this process, the expert is able to accomplish this task with precision and exactitude. There does however exist a significant series of diagnostic problems, for which simple visual analysis of the image is insufficient for the specific pathologic characterization of tissues. It was this clinical requirement which encouraged scientists to develop advanced methodologies concerning both acquisition systems and image processing methods. As a starting point, researchers naturally considered the image characteristics that radiologists use explicitly or implicitly in their evaluation of tissue appearance. Intensity, morphology and texture are usually quoted as important characteristics. Image texture is known to be a particularly sensitive characteristic in the evaluation of pathologies. Its visual analysis has always been used in medical imaging in order to establish a diagnosis. It is however well known that the human observer has only a limited sensitivity to textural properties, whereas mathematical techniques for texture analysis give quantitative and therefore objective elements. It is for this reason that many studies have been carried out in the aim of automating this analysis. This chapter is composed of an introduction, a conclusion and three parts : part two outlines the main methods of texture analysis in the medical field, part three summarizes the medical applications which use these techniques and part four describes several experiments in greater detail.

This chapter deals with a semi-automatic approach to the processing of images and to the extraction of regions and volumes of interest. Moreover, it describes the two methods of isovolume segmentation and pseudo-3-D segmentation, which have been installed in different European hospitals for evaluation purposes. Emphasis is placed on segmentation of spatial sequences, and on medical user's requirements: processing speed, ease of use, robustness, transparency, etc.

This paper presents a new approach for providing consistent anatomical structure models based on a geometrical description and on topological features. Volumetric and surfacic techniques are combined to make the best use of the knowledge we have about the object.

The reconstruction process begins with the location of the object in the digital volume. According to the geometrical continuity, active ray tracing in association with geometrical and textural knowledge allows for an efficient interactive investigation.

This investigation gives a rough shape of the object, from which a skeleton is extracted inducing the basis for a geometrical clustering. This clustering makes it possible to get an efficient geometrical characterization of each elementary part of the object.

A first model is then produced, according to the topological information gathered through the previous step. This information makes it possible to extract relevant points from which a triangulated surface can be efficiently reconstructed.

The modeling process is done by turning the triangulated surface into a free-form surface, which is corrected with very local adjustments through a fine texture analysis.

A short historical review of facial measurements and computer graphics leading to visualisation systems is given. The use of a computer graphics workstation developed at University College London for the simulation and planning of maxillofacial and cranial surgery is described. This creates a display of anatomical surfaces from three dimensional (3-D) datasets generated by Computerised Tomography, Magnetic Resonance, Ultrasound medical imaging systems and by a purpose built optical body scanner. The workstation offers a wide variety of facilities allowing the operator to interact with and interrogate these three dimensional datasets via the displayed images. The manner in which the software allows the operator to plan surgical procedures and to produce surgical models for rehearsal and prosthesis manufacture is discussed and illustrated with case material. The use of the system for studying facial form, growth and change is discussed, and clinical, morphological and forensic applications are described.

Hinging on approaches from the fields of clinical imaging and information technology assessment in medicine, a validation scheme is presented for 3-D biomedical image interpretation. This scheme considers measurement of algorithmic accuracy and evaluation of clinical functionality and usefulness. The former component is meant to characterize the intrinsic quality of a method, the latter to assess how well a method serves its clininal goals. Four examples from our own research are used to illustrate this scheme. We report on the validation of (1) a semi-automatic method to quantify myocardial perfusion on SPECT images, (2) bone segmentation accuracy on spiral CT images, (3) a workstation for stereotactic neurosurgery, and, (4) stereolithographic models.