Where medical images are concerned, appropriate grounds for image processing are sometimes misunderstood and such efforts should properly be divided into the separate objectives of: 1) image reconstruction; 2) image understanding and 3) image processing.

Two different types of methods for acquiring MR data more quickly have been explored here. The first set of methods uses half the normal number of measurements and is capable of shortening acquisition times by a factor of two. The second type of method uses measurements from multiple echoes to synthesize a 'conventional' spin-warp raw data set. This second method is capable of shortening data acquisition times by factors of four to eight. Three types of 'half the data' methods have been tested. The first uses the central half of the phase encoding steps, and reconstructs on the usual 256 x 256 matrix using a sinc interpolation; with this first method, signal to noise ratio is improved, but spatial resolution in the phase encoding direction is reduced. The second method uses all the upper (or lower) half of the normal phase encoding steps, places them in an otherwise empty 256 x 256 matrix, reconstructs, makes a phase correction, takes the real part. This 'half Fourier' method uses the phase information to retain full resolution with only half the data (really 53%) but at the cost of a reduction of signal to noise ratio. for the third method, every second phase encoding measurement is taken, along with eight extras in the center of the data space, empty data lines are filled by cubic interpolation, and a FFT recontruction is done. For this method, spatial resolution and signal to noise ratio are maintained, but two 'ghost' images (low amplitude-occur because of the inherent undersampling of this method. The 'multiple echo' method for faster data acquisition makes use of a sequence that acquires different phase encoding steps in the data from different echoes. the best strategy is to take the central (low frequency) measurements from the lowest numbered echoes, and the higher frequency measurements from the later echoes. The practical details of this method include correcting position, phase, and amplitude of measurements from different echoes. It is also important to use data acquisition sequences that are very well adjusted refered to stimulated echoes and eddy currents. The result is that data can be acquired four to eight times faster than normal for spin-warp imaging, almost full resolution is retained, and signal to noise ratio is reduced somewhat.

The basic task of making separate images of the two largest peaks of the MR proton spectrum (water and fat) has been approached as a problem of pixel by pixel spectroscopy and as a phase correction problem wherein one encodes information about frequency into the phase angle of a complex MR image. Images of induced magnRtic field (susceptibility) can also be made as part of the fat/water image calculations.

A modified Fan Beam reconstruction algorithm is presented that takes into account the delays introduced by multiplexing in CT's with continuous X-ray emission. The modified solution appears to be instrumentable in very much the same way as the classic Filtered Backprojection algorithm and yields an improved image quality.

For real-time analysis of light microscopic images by using structural features we have developed a special image processing system. The major components of the system are a parallel multi--microprocessor system, an image buffer, a stepping motor driver module for x-, y-, and z-movement of the microscope, a video processing module and a binary image module. The two latter modules are special hardware to generate structural parameters of the video image in real-time. The parameters are derived from the number and optical density of individual objects within each image line. An object is recognized if its optical density exceeds a predefined threshold and its size falls into a programmable interval. We have applied this system to objectively evaluate histological changes occuring in neoplasias treated preoperatively with chemotherapy. The result is an automated mapping of residual tumor lesions for each specimen. These maps are used to define chemotherapy effects. It should be emphasized that because of its flexibility, the system described can be used more generally to analyse structures of histological images.

For several years, the formation and evolution of thrombi in small arteries of rats has been quantitatively studied at the Laboratory of Physiology and Physiopathology at the V.U.B. Global size parameters can be determined by projecting the image of a small arterial segment onto photosensitive cells. The transmitted light intensity is a measure for the thrombotic phenomenon. This unique method permitted extensive in vivo study of the platelet-vessel wall interaction and local thrombosis. Now, a further development has emerged with the aim to improve the resolution of these measurements in order to get information on texture and form of the thrombotic mass at any stage of its evolution. Therefore a thorough understanding of how light propagates through non hemolized blood was essential. For this purpose, the Medical Informatics department developed a system to record and process digital images of the thrombotic phenomenon. For the processing and attempt to reconstruct the thrombi, a model describing the light transmission in a dispersive medium such as flowing blood had to be worked out. Application of results from Twersky's multiple scattering theory, combined with appropriate border conditions and parameter values was attempted. In the particular situation we studied, the dispersive properties of the flowing blood were found to be highly anisotropic. An explanation for this phenomenon could be given by considering the alignment of red blood cells in the blood flow. In order to explain the measured intensity profiles, we had to postulate alignment in the plane perpendicular to the flow as well. The theoretical predictions are in good agreement with the experimental values if we assume almost perfect alignment of the erythrocytes such that their short axes are pointing in the direction of the center of the artery. Conclusive evidence of the interaction between local flow properties and light transmission could be found by observing arteries with perturbated flow.

We describe the use of digital deconvolution in the study of muscle striations below the resolution imposed by the optical diffraction of our video light microscope. To use deconvolution procedures on muscle images, the transfer function of the optical system is first characterized. This is accomplished by imaging a step object and fitting the image with the combination of a first order Bessel and Gaussian function using a non-linear least squares approach. Due to the ill-conditioned nature of deconvolution, however, ambiguity in the reconstruction is sometimes found. To allow better estimation of the true object, separate deconvolution approaches are used and the reconstructions compared. In this manner, the fine structure of muscle striations is determined.

A set of computer-assisted procedures was developed for sampling and processing quantitative information from electron images. Data are manually collected using a digitizing tablet, and subsequently elaborated with a I B M- compatible personal computer. Programs presented here deals with stereological calculations, numerical taxonomy and basic statistics applied to tumor cell samples.

In image-forming optical systems the image of a three-dimensional object consists of a superposition of focused and defocused object layers. For a quantitative evaluation of the object it is necessary to decompose the superposition image into different images corresponding to single object layers. For this purpose the object radiation is measured with different optical transfer functions of the imaging system, for example by simply changing the focus plane. Each image contains focused and defocused parts of the object and can be described as a linear equation of the object layers, assuming linear space-invariant imaging properties. From these images the real object distribution can be calculated by the evaluation of the resulting linear system of equations in the Fourier domain. Due to noise in the detected images it is only possible to get an estimate of the true object distribution. In our case this estimate is based on an integral minimal mean square error in the reconstructed object. The algorithm is presented and demonstrated by simulation experiments and reconstructions of real human cell images in optical microscopy.

Quantitative evaluation of native biologic cell material can be performed regarding the absorption in ultraviolet light at several specific wavelengths. Due to motion of unfixed cells consecutive images of the cells may be shifted, rotated, and even shrinked. For motion vector analysis each pixel of the cell image sequence is described by a multidimensional feature vector containing local neighbourhood relations. Based on this vector set, the motion vectors are determined by correlation techniques and euclidian distance measurements. The global movement parameters are evaluated from the calculated local motion vectors to compensate the movement of cells between consecutive image scenes. The procedure is presented and demonstrated by experiments for separation of nucleic acid specific and nuclear protein specific structures in unfixed and unstained clinical cell material.

A system for the production of a four dimensional (moving three dimensional) human epicardial left ventriculogram, modelled and highlighted to show regional wall motion changes, is described. The moving image is derived by fitting a surface to the three dimensional coordinates of coronary artery bifurcations. These are determined by analysis of digitised biplane coronary cineangiograms. This image system not only provides a unique 3-D view of left ventricular activity but might also provide measures of cardiac dynamics such as, stroke volume and velocity of wall movement. The system is not fully automated although operator interaction may be minimised. Further work on vessel tracking systems is required before full automation is possible.

imaging modalities such as X-Ray computerized Tomography (CT), Nuclear Medecine and Nuclear Magnetic Resonance can produce three-dimensional (3-D) arrays of numerical data of medical object internal structures. The analysis of 3-D data by synthetic generation of realistic images is an important area of computer graphics and imaging.

Flexible image presentations of anatomic structures are now routine practice for CT diagnosis. Saggital, coronal, para-axial, and oblique views reformatted from regularly spaced CT images are now a standard software package for all CT manufacturers. These images are unrestricted by the orientation of data acquisition where only flat planes of anatomy are scanned. As this viewing flexibility improves, image orientations are suggested that are further liberated from orthogonal image storage architectures of past work. Image orientation for those views mentioned above are all planar. This paper illustrates a digital image that is three-dimensional in its description but flat in its presentation - an image that follows the structural shape of anatomy. More simply, we present a generalized "curved-surface" image and demonstrate its clinical utility in examinations of the mandible, lumbar and cervical spine. We show by examples given here, how clinical diagnosis is aided and three-dimensional structure is more easily understood when researchers, clinicians and surgeons visualize anatomy presentations that follow anatomic curvature.

In this paper we describe some of our experiments in the processing and display of 3-D images from CT-scanners. In the first part of the paper we describe the implemented surface displays and in the second part we discuss some applications of 3-D image transformations.

Two methods for projecting 3D CT data onto plane X-ray radiographs for the use in stereotactic neurosurgery are discussed. Both methods deal with completely arbitrary projection configurations and do not require additional computer equipment. Simulation tests have proven the reliability and necessary accuracy for operational purposes.

Computer graphics has many forms. When applied in medicine, it can range from simple two dimensional charts and graphs to rendering of three-dimensional scenes. Computer graphic displays of molecular or large anatomic structures have been used to great advantage by numerous medical researchers. In addition, graphic presentations can be dynamic where displays are controlled by physician-user commands, or the presentations can be static, where views are recorded in discrete frames for later distribution or permanent archival. In medicine both interactive and static forms of computer graphics have their proper place in the effective delivery of health care. Computer graphics, however, changes constantly in the area of software techniques, hardware improvements and its clinical application. What may be medically appropriate today in the use of computer graphics can soon become inadequate and well behind the new advances that so quickly follow. In this paper the key feature of computer communication is discussed that aids in the clinical utility of computer graphics in medicine. It is distribution. Distribution in terms of instantaneous computer graphic software updates and more importantly, distribution of meaningful three-dimensional presentations to referring physicians. Physicians who, working in their private offices, have no routine access to medical work stations. In this environment three dimensional presentations of anatomy are static in nature, but must deliver realistic views of critical structures. This paper outlines how computer communication provides the essential ingredient to the provision of this service. As an illustration, the electronic distribution of software to generate three dimensional views of complex anatomoic structures is discussed. Sample views are included.

The desirability of an integrated (digital) communication system for medical images is widely accepted. In the USA and in Europe several experimental projects are in progress to realise (a part of) such a system. Among these is the IMAGIS project in the Netherlands. From the conclusions of the preliminary studies performed, some requirements can be formulated such a system should meet in order to be accepted by its users. 1. The storage resolution of the images should match the maximum resolution of the presently acquired digital images. This determines the amount of data and therefore the storage requirements. 2. The desired images should be there when needed. This time constraint determines the speed requirements to be imposed on the system. As compared to current standards, very large storage capacities and very fast communication media are needed to meet these requirements. By employing cacheing techniques and suitable data compression schemes for the storage and by carefully choosing the network protocols, bare capacity demands can be alleviated. A communication network is needed to make the imaging system available over a larger area. As the network is very likely to become a major bottleneck for system performance, effects of variation of various attributes have to be carefully studied and analysed. After interesting results had been obtained (although preliminary) using a simulation model for a layered storage structure, it was decided to apply simulation also to this problem. Effects of network topology, access protocols and buffering strategies will be tested. Changes in performance resulting from changes in various network parameters will be studied. Results of this study at its present state are presented.

In most application of diagnostic imaging the amount of digitized information (up to 4096 grey levels per pixel) cannot be detected by the human observer as a whole, so that it is common practice to interactively select a suitable intensity window for the display, sometimes with saturation effects. To overcome this problem many different contrast stretching approaches have been proposed, to almost equalize the diagnostic information on the image. The solution proposed here is based on a non-linear transformation, acting as an adaptive histogram equalization of the local differences, in the hypothesis of gaussian distributions. The obtained results are closely related to other enhancement methods, with some advantage in the look-up filter implementation. The computational cost of the realization is evaluated, as well as the loss of performance determined by suboptimal block-processing and bilinear interpolation. Examples are referred in the contrast enhancement of C.T. images and digitized X-Rays.

This paper presents an approach to image understanding focusing on low level image pro-cessing and proposes a rule-based approach as part of larger knowledge-based system. The general system has a yerarchical structure that comprises several knowledge-based layers. The main idea is to confine at the lower level the domain independent knowledge and to reserve the higher levels for the domain dependent knowledge, that is for the interpretation.

The Discrete Cosine Transform (DCT) is recognized as an important tool for image compression techniques. Its use in image restoration is, however, not well known. It is the aim of this paper to provide a restoration method for a sequence of images using the DCT as well for the deblurring as for the noise reduction. It is shown that the DCT can play an interesting role in the deconvolution problem for linear imaging systems with finite, invariant and symmetric impulse response. It is further shown that the noise reduction can be performed onto an image sequence using a time adaptive Kalman filter in the domain of the Karhunen-Loeve transform which is approximated by the DCT.

The major aims in tissue characterization are the differentiation between normal, benign and malignant (primary/metastasic) tissues. Usually the information of standard spin echo images as well as the MR parameters T1, T2 and spin density are used for clinical diagnosis. Pattern recognition methods are applied to improve the specificity of MRI. The variety of all these parameters requires the selection of the parameters containing the relevant diagnostic information. Therefore images with optimum contrast to a given diagnostic task are generated.

Computerized analysis of 2-D gel electrophoretic images is the basic approach to solve various problems involved in the interpretation of the results of this technique. In this paper many aspects are investigated: smoothing procedures, relation between protein quantities and integrated optical densities, segmentation techniques, and wide feature extraction, pointing out also the importance of spatial resolution. In particular, a newly developed segmentation technique is applied to the smoothed images and the promising results are discussed.

The work described was originally aimed at providing a new diagnostic technique for the early detection of malignant ocular tumors through their spectral signature. The instrument developed comprises a modified fundus camera, a Charge-coupled device (CCD) camera and a 16 bit microcomputer equipped with floppy disk drives and a 512 x 512X 8-bit display device. The system allows the recording of digitized fundus or iris reflectance pictures in eight spec-tral bands between 500 and 1100 nm. After calibration and preprocessing of the data, a multi-spectral analysis is performed by means of a VAX computer. The image processing methods are described and their ability to characterize pigmented lesions or other ocular anatomical fea-tures through their spectral signature is evaluated.