(This is almost verbatim the text of the conference presented in plenary session PLO6 at the E.C.O. 2 symposium in Paris on april 27, 1989.) I would like to devote this talk to a tutorial review of the domain of optical computing as seen from the viewpoint of optical interconnects. We shall start with a short critical assessment of the state of optical computing, showing that optical interconnects are an important part of it. We shall then review the technologies and the families of components available and whose continuing development appears necessary in this context. The question of interconnects in microelectronics will then be examined, clearly supporting the assertion that optics has a part to play there. We shall close with a review of the fundamental advantages that can reasonably be expected from the introduction of optical interconnects in electronic computers. The general ideas will be illustrated whenever possible by recent examples from the litterature or from projects presently under investigation in our Institute in Orsay. To a large extent, this presentation will elaborate on work inspired by the pionneering 1984 article by Goodman, Leonberger, Kung and Athale.
With proper calibration several image acquisition modalities produce images whose intensity is proportional to the density of some conserved quantity or to the projection of such a density. Such "density" images may be obtained for example in computer assisted tomography or magnetic resonance imaging (MRI); projected density images may be obtained via radiography or scintigraphy. Within a given imaging modality, density images of a given set of objects are related through one-to-one geometrical transformations. Differences among the images may arise from object motion between acquisitions or from changes in the acquisition device itself. In this paper techniques are presented for finding approximate geometrical transformations between density images of a set of objects acquired via the same modality. Applications are discussed with respect to motion artifacts that obscure artery images in digital subtraction angiography (DSA) and field inhomogeniety artifacts in MRI. Experimental results in DSA are presented to demonstrate that the principle difficulty lies not in finding an effective transformation for removing motion artifacts, but in selecting one that distinguishes between the artifacts and the artery.
Rubber Sheet Masking (RSM) is a fully automated method for reducing motion artefacts in Digital Subtraction Angiography (DSA). RSM is based on Venot's registration method for DSA images but improves this method in two respects. Firstly, local image statistics are better taken into account. Secondly, RSM does not require user interaction to define region of interest for motion registration. Attention has furthermore been paid to fast implementation of the method, and to its clinical usefulness. The conclusion of the study is that RSM compares favourably with the conventional interactive pixel shift methods.
Image sequences of moving organs provide pattern motion informations and their appropriate management leads to new looks at image segmentation. We propose in this paper an illustration of the benefits that can be expected for coronary vessels analysis. A new method is described which combines the motion estimation with the frame-to-frame structure detection in a natural way such that they act interactively. The first step consists of the extraction of the vessel centerlines in one image. These data are further organized in meaningful constituents or branches of the coronary arterial tree. Then, the motion is estimated at each point of the centerlines through a gradient-based method. These motion estimates supply an initial positioning of an active contour model (or "snake") in the next image. This model adapts itself by changing its shape and locks accurately onto the new distorted centerlines. This whole process is then reiterated on the subsequent images to depict the dynamic behaviour of all the relevant branches. The main interests of this scheme are : (1) the snakes operate locally, so a fast detection can be performed; (2) the skeleton extraction is fully guided by the confluence of the motion estimation and the active contour modelling; (3) it can be easily extended to derive the boundaries of the coronary arteries; (4) both morphological and kinetic properties are achieved on a quantitative basis.
Factor Analysis of Dynamic Structures (FADS) has been developed since ten years in the medical imaging field to process dynamic image sequences. Its aim is to estimate the different underlying physiological mechanisms. This paper summarises the conventional FADS algorithm and discusses its limitations. A priori information has been used by several authors in order to improve the estimations. After a critical review of these approaches, we propose an original way of introducing constraints, which is based on the a priori knowledge of the physiological data. It is applied to the study of 99mTc HMDP early kinetics in osseous tissues.
We propose a new method of three dimensional image reconstruction from a limited set of projection data, which uses a decomposition of the object distribution on "constrained natural voxels". This method combines the advantages of the natural voxel decomposition, which matches exactly the measurements recording, and those of constrained algorithms, which incorporate a priori informations in the reconstruction problem. The natural elements are the elementary integration volumes, covered by the beam paths. We propose in this method to weight each natural element with a function that can translate a priori informations on support and density. We have applied this method to two 3D recording systems used in Nuclear Medicine : the slanted hole collimator and the rotating slit collimator.
We address the problem of reconstructing a 3-D object using a few of its conic projections measured with a 2-D X-ray detector. This problem is ill-posed and prior information must therefore be used to regularize the solution. We propose a non-parametric method based on a detection-estimation scheme that is particulary well-suited to the reconstruction of sparse objects such as 3-D vascular structures. We present results of 3-D reconstruction obtained from both computer simulated and experimentally measured projections.
In this paper a new ultrasound tomographic image algorithm is presented. A complete laboratory system is built up to test the algorithm in experimental conditions. The proposed system is based on a physical model consisting of a bidimensional distribution of single scattering elements. Multiple scattering is neglected, so Born approximation is assumed. This tomographic technique only requires two orthogonal scanning sections. For each rotational position of the object, data are collected by means of the complete data set method in transmission mode. After a numeric envelope detection, the received signals are back-projected in the space-domain through a scalar function. The reconstruction of each scattering element is accomplished by correlating the ultrasound time of flight and attenuation with the points' loci given by the possible positions of the scattering element. The points' locus is represented by an ellipse with the focuses located on the transmitter and receiver positions. In the image matrix the ellipses' contributions are coherently summed in the position of the scattering element. Computer simulations of cylindrical-shaped objects have pointed out the performances of the reconstruction algorithm. Preliminary experimental results show the laboratory system features. On the basis of these results an experimental procedure to test the confidence and repeatability of ultrasonic measurements on human carotid vessel is proposed.
The first part of this article is a brief presentation of the models selected to represent the current-voltage relationships in biological tissue with a view to their application in impedance tomography. The second and third parts develop the principal resolution methods for these models: - The analytical methods (above all the Barber and Brown method (7-14)). - The digital methods (Kim (1), Wexler (20), Alessandrini (17). Some remarks and comments concerning implementation of these reconstruction algorithms will be made in conclusion to this presentation.
A multi-slice Positron Camera organizes data as a set of parallel slices whose thickness is defined by the axial response function (ARF) of the tomograph. This ARF is usually not uniform and the sensitivity is maximum for a point source located at the center of the slice. As a consequence, the partial volume effect (quantitation loss for structures that are smaller than two to three times the FWHM resolution) will he dependent not only on the size of the structure but also on its relative position with regard to the ARF. We have developed a computer simulation to analyze this effect. Objects of different size and shape are examined and it is shown that on a system with a 6 mm axial resolution (FWHM), a shift of 3 mm can introduce more than 10% variation on the final measurement. The simulation is confirmed by experimental results on a high resolution PET. We conclude that positioning and repositioning must be handled with great care in order to make reproducible measurements and allow accurate partial volume corrections.
In MR imaging, the extent of the acquired spatial frequencies of the object is necessarily finite. The resulting image shows artefacts caused by "truncation" of its Fourier components. These are known as Gibbs artefacts or ringing artefacts. These artefacts are particularly. visible when the time-saving reduced acquisition method is used, say, when scanning only the lowest 70% of the 256 data lines. Filtering the data results in loss of resolution. A method is described that estimates the high frequency data from the low-frequency data lines, with the likelihood of the image as criterion. It is a computationally very efficient method, since it requires practically only two extra Fourier transforms, in addition to the normal. reconstruction. The results of this method on MR images of human subjects are promising. Evaluations on a 70% acquisition image show about 20% decrease of the error energy after processing. "Error energy" is defined as the total power of the difference to a 256-data-lines reference image. The elimination of ringing artefacts then appears almost complete..
Volume rendering not only produces high quality 3D images of single organs directly from the intensity data in CT, MR, SPECT, PET, and ultrasound images, but it also can be used to show the relationship among multiple anatomic, treatment, and image objects. In this paper we will explain methods, show results, and discuss the effectiveness of 1. simultaneously volume rendering multiple organs from a single image data set, using transparency and color; 2. rendering polygonally defined treatment objects such as prostheses and radiation treatment beams with volume rendered anatomy. We will also discuss ongoing work in 1. rendering multiple image sets, e.g., radiation dose distributions and medical image data, into a single image via volume rendering, e.g., to show the relation between isodose surfaces and anatomic objects; 2. texture mapping grey scale slices onto clipping planes on volume rendered anatomy, to show the relationship between physiological and anatomic data, or to show the subtle anatomic intensity variations in its 3D context.
The morphology of anatomical objects from MRI scans or X-ray CT scans can be represented efficiently using triangulated surfaces. A triangulated surface is a closed surface (consisting in triangles) which is computed from polygons in a sequence of planes. The polygons interpolate object boundaries in a sequence of MR or CT slices. We present a new algorithm to create such a triangulation which overcomes many of the currently existing restrictions for these algorithms. It automatically generates a triangulated surface from arbitrary sets of closed polygons. The surface generation is initiated by a linear interpolation of the object's boundary in a set of parallel planes intersecting the object polygons. Object boundaries in the intersection planes and in the scan sequence form a grid which is used for the final triangulation. It can be shown that the approximation will be nearly optimal with respect to the input information. Slices, in which objects are outlined, may be curved and need not be parallel to each other. Results of the automatic reconstruction of anatomical objects from cranial CT scans and MR scans are presented.
In biomedical engineering many applications (C.A.T., R.M.N., microscopy) involve the representation of 3D numeric data to help in the comprehension of objects structure. In this paper we are interested in the case where the 3D object information is obtained from a sequence of 2D cross-sectional images. The proposed method is the polyhedral approximation of an object's surface by a set of triangular planar patches what is called triangulation. Although this approach has been often developed, it suffers some limitations, for instance when the contours are very distorted or when there is more than one contour by slice (multiple-contour). We propose a triangulation algorithm which provides a solution to these problems and which can be used even for complex shape objects. We present the application of our algorithm to the 3D representation of three real biomedical objects of increasing complexity.
A convenient form of 3-D object representation for medical data is the voxel approach. Both simple and complex objects can be represented easily, but large numbers of voxels are required to describe smooth shapes. Considerable data compression can be achieved by representing only surface voxels. Enumeration of these in an ordered back-to-front sequence generates a list of surface voxels which incorporates shape information. Simple traversal of this ordered surface list corresponds to a full back-to-front traversal of the surface elements of the original object, which essentially solves the hidden surface problem. For cubic voxels only eight ordered surface lists are required to generate hidden surface images from any angle. This approach has been applied to radionuclide tomograms sampled at 643 resolution, achieving real time display using a conventional medical imaging computer. The technique has also been applied to CT and MRI data sampled at higher resolution. Segmentation and traversal of an N3 array of voxels to obtain surface lists is an 0(N3) procedure. However, list length is proportional to object surface area, and hence subsequent display is only an 0(N2) procedure. This property enables interactive display of large arrays without using special purpose hardware.
A display method is proposed in which the relaxation times T1 and T2, and the proton density of tissue in each pixel in an MR image are simultaneously expressed in color features in a unified way. MR images were made from phantoms, volunteers and patients in such a way that T1 and T2 and proton density images could be calculated. Color images were computed from these images using matrix multiplication on a pixel base. In this way the color combination in each pixel represents the properties of that particular pixel by a unique mixing of the elementary colors red, green and blue. Color resolution could be modified using different choices of a reference color triangle in which the color combinations were defined. This method of representation offers a means for displaying multiple features as T1 and T2 in one directly interpretable image, independent of instrumental settings.
Within the field of radiology, assistance with computer and communication systems may be applied to generation, storing, transmission, viewing, analyzing and interpreting of images. As a result, digital image management and communication systems will be applied at various levels in the health care system. Four groups of people are somehow involved or affected by this process. These are, first of all, the patients and the medical personnel, but also the scientific-engineering community and the group of professions involved with financing and/or administering these systems. Each group approaches computer assisted radiology from a particular point of view. The paper outlines some aspects as regards the different perceptions of these groups, which need to be clarified in order to successfully realise computer assisted radiology.
The processing of medical images requires the handling of complex structured sets of elementary objects (images, curves,... and their associated parameters). Usually, an elementary object cannot be interpreted without information concerning the structure to which it belongs (e.g image sequences). Then it is necessary to consider the whole structure like an atomic semantic entity, object of an image data base. As specific tools are necessary to manage these objects, an object oriented handling system (OHS), part of our medical image data base project (BDIM), was developed to perform : i) the array storage management, ii) the interface between applications and the BDIM to have access to objects (create, update, delete...) and components (navigation inside object structures, access to arrays and parameters). The image handling system (IHS) decribed here is the user level part of the OHS. IHS allows the evolution of the data base environment by adding or updating acquisition and/or processing functionalities. To unify data access methods, the concept of logical file is introduced as a special class of BDIM objects. The logical file does not necessitate the use of a specific declaration for the different kinds of images because it is possible, for a desired processing , to have access to the only concerned data.
The successful implementation of medical PACS requires the use of powerful compression coding to archive and transmit images and image series. In this prospect, we developed and evaluated a general irreversible method to compress medical image series achieving compression ratios as high as 50:1 to 100:1. The implemented technique combines two coding schemes which multiply their effects. A principal component analysis, first step of the conventional factor analysis of dynamic structures (FADS), is applied to the original dynamic series. A limited number of principal components (curves) and their associated spatial distribution (images) are computed. Then, each image is compressed by using an adaptive block-quantization technique computing the 2D discrete cosine transform (DCT). To reconstruct the images associated to the principal components, an inverse DCT is applied. Then, the original series is computed from the reconstructed images combined with the principal components, stored without any modification. The reconstructed series is compared to the original one, as well as the time-activity curves generated using different regions of interests (ROI) and the results obtained applying FADS to the two series.
In this paper several reversible (error-free) decorrelation schemes for time series of images are described. Distinction is made between spatial-temporal schemes versus temporal-spatial schemes, motion compensating versus non compensating schemes, and between temporal extrapolation versus temporal interpolation methods. Simulation on three series of angiograms shows that reversible interframe decorrelation offers no significant improvement over intraframe decorrelation by means of Hierarchical Interpolation. For a sequence of video-conferencing images interframe decorrelation did yield improved performance over intraframe HINT. However, this type of images is usually compressed irreversibly.
There appears to be considerable value in creating an expert system capable of handling medical images. It is suggested that the best way of implementing such a system is to use an expert system shell beneath which image processing primitives run. However, a number of different models can be used to represent such a system, depending primarily on the way in which the user interacts with the system and on the amount of feedback between the different components. These are discussed in some detail, and the corresponding aims identified. Although few substantial systems have been realized, a series of potential application areas are presented. These are those associated with feature extraction, image interpretation, the use of image databases, query languages, and clinical assistance and teaching systems. An interesting example of such a system is illustrated in the problems associated with the identification of common structures needed to generate a 3-D image from a bi-plane angiogram. Finally, some indication of methods for implementing such system, and the problems likely to be encountered are given.
At present, the phase and amplitude images are used in nuclear medicine to aid the diagnosis of cardiac regional wall motion abnormalities (RWMA). These images contain only part of the information present in the original radionuclide images, and have to be mentally integrated with other information to obtain a diagnosis. The proposed expert images offer a direct diagnosis in terms of eight degrees of RWMA of cardiac left ventricle from normal to dyskinetic. They are based on a multiparametric description of the pixel time behavior in equilibrium gated studies, on image filtering, and signal, prediction and pattern analyses: a physician "teaches" the computer by drawing boundaries of domains of regional wall motion abnormality classes into images of well defined cases, and the computer then emulates on new, unknown cases diagnostic drawings as they would be drawn by the physicians. The proposed expert images represent a step beyond the categories of present functional images.
X-ray mammogram is the only breast cancer detection technique presently available with proven efficacy and nearly 50% of breast carcinomas demonstrate microcalcifications on mammograms. Moreover, it has been shown that clustered microcalcifications where the only sign of abnormality in about 36% of nonpalpable, clinically occult breast cancers.
In a clinical pilot study radiofrequency echograms were acquired during routine echography of intraocular tumours. The data acquisition was performed with an "add-on" device, which was developed at our laboratory. The acquired data were pre-processed to remove the effects due to equipment performance and beam characteristics. The analysis was performed in the frequency domain (acoustospectrography) and acoustic tissue parameters like the attentuation coefficient and backscatter cross section estimated. In addition the gray level statistics of B-mode images were analysed. A set of parameters was thus obtained which was subjected to a discriminant analysis, based on a classification of the tumours by histology (i.e. after removal of the eyes). The results show a significant discriminability between different tumours and even between histological types of the same tumour (choroidal melanoma).
Synthetic MR images that enhance tissue related contrast are proposed and compared. They are based on a subset supervised subspace methods. Advantages and weaknesses of individual transforms are discussed, including the discriminatory power, contrast and contrast/noise of the resulting images. For example, it is shown that eigenimages can be derived in a simple way without eigenanalysis, that eigenfilter is inferior to the maximal contrast transform, and that the common weakness of both of these methods can be compensated for by a blending transform. Examples of the corresponding synthetic images are provided as well.
Studies of human and ideal observer decision performance are designed to estimate a measure of observer performance (e.g. efficiency or d'). The experimenter would like to use a relatively small number of images or decision trials because of a variety of constraints (observer time, image preparation time etc.). Sampling variance results will be presented and experimental strategies for both ROC and Forced Choice studies will be analyzed and discussed. In many situations ROC rating scale experiments are the most economical.
A PET camera consisting of two pairs of parallel area detectors has been installed at the cyclotron unit of VUB. The detectors are High Density Avalanche Chambers (HIDAC) wire-chambers with a stack of 4 or 6 lead gamma-electron converters, the sensitive area being 30 by 30 cm. The detectors are mounted on a commercial gantry allowing a 180 degree rotation during acquisition, as needed for a fully 3D image reconstruction. The camera has been interfaced to a token-ring computer network consisting of 5 workstations among which the various tasks (acquisition, reconstruction, display) can be distributed. Each coincident event is coded in 48 bits and is transmitted to the computer bus via a 512 kbytes dual ported buffer memory allowing data rates of up to 50 kHz. Fully 3D image reconstruction software has been developed, and includes new reconstruction algorithms allowing a better utilization of the available projection data. Preliminary measurements and imaging of phantoms and small animals (with 18FDG) have been performed with two of the four detectors mounted on the gantry. They indicate the expected 3D isotropic spatial resolution of 3.5 mm (FWHM, line source in air) and a sensitivity of 4 cps/μCi for a centred point source in air, corresponding to typical data rates of a few kHz. This latter figure is expected to improve by a factor of 4 after coupling of the second detector pair, since the coincidence sensitivity of this second detector pair is a factor 3 higher than that of the first one.
We describe a new method for the assessment of the bone mineral content of the peripheral skeleton. The method makes use of a single photon counting and imaging area detector, the Multi-Wire Proportional Chamber (MWPC). The application of this system to the measurement of the Bone Mineral Density (BMD) of calcaneum, using the MWPC as a dual photon absorptiometer with the isotopes pair 1251, 244,m, is discussed.
Aim of the work is the visualization of coronary arteries down to 1 mm diameter with an iodine mass density of 1 mg/cm , thus allowing non-invasive investigations by intravenous injection of the contrast agent. Digital Subtraction Angiography (DSA) in energy subtraction mode (dichromography) is employed for this purpose. The two images Cor subtraction are taken at photon energies just below and above the iodine K-edge (33.17 keV). After subtraction the background contrast - such as bone and soft tissue - is suppressed and the iodinated structures are strongly enhanced because of the abrupt change of absorption at the edge. The two monoenergetic beams (bandwidth about 250 eV) with high intensity (about 1011 photons/mm /s) are only available if synchrotron radiation is used. In HASYLAB at DESY (Hamburg, FRG) the system NIKOS was developed for dichromography. It consists of six main parts: A wiggler beam line, a monochromator which filters the two 12 cm wide beams out of the white synchrotron radiation beam, a fast scanning device, a fast low-noise two-line detector, a safety system and a computer system. At present, one scan (two images) lasts 1 s. The images from the in-vivo investigations of dogs have been promising. The right coronary artery (diameter 1.5 mm) was clearly visible.
Tomographic processes for biomedical purposes has observed a fantastic development during the last few years, such as ultrasonic echography, X-ray tomodensitometry or NMR tomography and more recently active microwave imaging. This new modality offers a low spatial reslution, depending of the frequency, but, in some cases significant contrasts related to dielectric properties of tissues. The interpretation of microwave images presents some specific aspects related to the diffraction mode of interaction between microwave beams and tissues. Tomographic algotithms have been developped to compensate for such diffraction effects. More works are still needed to improve image quality in terms of spatial resolution or artefact delete. Initially, motivated by non-invasive thermal control of deep hyperthermia treatment, experiments have been conducted on phantoms with different heating modalities. This paper presents some results to evaluate spatial and thermal resolution capabilities of microwave tomography. More recently, to respect to contrast, more general applications have been investigated: detection of inflammatory processes, monitoring of lung water content changes,control of organs defrozing...Some prelimary results obtained with a 2.45 GHz planar microwave camera are presented.