Digital image processing in echocardiography means interactive analysis of a variety of information under different points of view. Repeated menu controlled analysis of images is very time consuming. As an alternative, a program construction system for dialogue oriented generation of specialized image processing programs may be used. The generated programs can solve the image processing tasks in a very effective way with minimum user interaction. This paper describes such a program construction system and illustrates the specification of single steps of analysis with the help of examples taken from the area of echocardiography.

This title is used also for an exhaustive overview of world-wide microscopic image analysis CPr8SJ. PRESTON and BPRTELS revie' not only the current state of the automation in cytological and histological image analysis but they present also a historical account of this field of computer aided image processing. bout 800 papers were referenced. mong them were nearly 20 papers by GDR authors published in the years 1973-82. The following paper will supplement the above mentioned extended jork in sense of *zooming - that means more details about a smaller region.

Image analysis offers the opportunity to analyse many processes in medicine, biology and engeneering in a quantitative manner. Experience shows that it is only by awareness of preparation methods and attention to software design that full benefit can be reaped from a picture processing system in the fields of cytology and histology. Some examples of special stains for automated analysis are given here and the effectiveness of commercially available software packages is investigated. The application of picture processing and development of related special hardware and software has been increasing within the last years. As PC-based picture processing systems can be purchased at reasonable costs more and more users are confronted with these problems. Experience shows that the quality of commercially available software packages differ and the requirements on the sample preparation needed for successful problem solutions are often underestimated. But as always, sample preparation is still the key to success in automated image analysis for cells and tissues. Hence, a problem solution requires the permanent interaction between sample preparation methods and algorithm development.

With the development and the increasing use of quantitative morphological techniques the concept of subvisual morphological properties has become the main topic of quantitative morphology in medicine especially in tumour pathology. Subvisual properties of cells and tissues summarize all that structural information of cells and tissues being present in a histological or cytological image but not being to recognize without special tools. As to extract that information the visual perception (also my means of a microscope) has to be substituted by the quantitative measurement of structures. The paper deals with the basic understanding of the " subvisuality" regarding the human seeing. As it will be shown almost all of the recent topics of image analysis with important clinical and theoretical impact can be reduced to the problem of subvisual morphological properties. Roughly 80 of all information we get by our brain are optical ones. During the evolution the eye-brain-system has developed the ability for the rapid and very adaptable recognition of highly complex patterns. Recently and in the foreseeable future this ability is not to substitute by any arteficial system. An essential characteristic of the human seeing is the high propertion of experience and learning processes being involved in the information processing mentioned above. Until now we know only small pieces of this kind of information processing practically nothing is known about " learning" in human pattern recognition. But if such

Caryoxnetry and histoinetry are the two characterization levels for histological patterns. A wide spectrum of caryoinetrical methods exists (characterization of size and form, densitometry, nuclear texture analysis) and there is often a good correlation between measured parameters and diagnosis as well as proosis, especially in tumor diagnostic /14/15/20/. The development of histoinetry has just begun.

For more than twenty five years it has been attempted to resolve the automated analysis of human chromosomes by machines. In this review about the application of image analysis techniques it will be tried to give a global overview about some important image proces sing and analysis steps in the automated chromosome analysis. The software is included into the commercially available slidebased analysis system SSCAN for IBMPC compatible computers. Chromosome analysis systems have been developed because of the demand of cyto geneticists for a reliable and economic routine method in human cytogenetit 1 aboratories.

In the context of the early recognition of cervical cancer the cytology has to detect tumor positive cases among a bulk of un suspicious specimens. Unfortunately there is a group of cases with a vague diaxosis, the so called PAP-ill-group. There are many reasons for the recent impossibility of a safe conclusion from the PAP-Ill-smears to their histological diagnoses (Pig. 1). Today for an exact statement a cone biopsy and a microscopical investigation of its histological sections for such caseS are necessary. That results in a high medical and economical expense and, last not least, that means a higher risk of complications and of future family planning problems for the women affected.

In contrast to cytophotomnetric studies of nuclear detail quantitative inorphomnetry of histologic specimens has only been developed to a limited degree as a diagnostic tool /13/. Counting of mnitoses is one of the oldest methods for quantification and has become a valuable predictor of good or poor prognosis of uterine soft tissue tumors /5/. In case of malignant melanoma the determination of tumor thickness has proven an important prognostic factor /4/. Measurements of endometrial gland volume enable the pathologist to discriminate endomnetrial carcinoma from hyperplasia /1/2/. Other applications of mnorphometry to special diagnostic problems could be added /13/. L

Computervision is used to overcome the mismatch between user models and implementation models of software systems for image analysis in nuclear medicine. Computer vision in nuclear medicine results in an active support of the user by the system. This is reached by modeling of imaging equipment and schedules scenes of interest and the process of visual image interpretation. Computer vision is demonstrated especially in the low level and medium level range. Special highlights are given for the estimation of image quality an uniform approach to enhancement and restoration of images and analysis of shape and dynamics of patterns. 1.

Bone scintigraphy is the most frequently used nuclearmedical procedure in oncological diagnostics. A method for the automatic identification and quantitition of bone lesions in bone scintigrams is described. The idea is: without an individual identification of the more than 100 distinguishable bones of the skeleton the scintigram quantitatively will be compared with eg. a ''normal''s scintigram. Before that procedure all the bones will be transformed in size and location to fit the measures of a standard skeleton. Grey levels remain untouched. After the quantitition of regional pulse contents sets of bone masks are used for the identification of the corresponding bones. The normalisation of pulses in bones and the elimination of background in the scintigram are unavoidable prerequisites. The automatic quantitative evaluation of bone scintigrams leads to an improvement of reproducibility accuracy and data presentation for the health care of cancer patients. Percentual deviations to normal left to right side and changes in time are localised and quantified in primary diagnostics follow up and therapy control. I .

For preoperative planning of dental implantations pictorial representations are required that permit to evaluate the size of teeth holes, size and structure of jaw bones, position of the mandibular channel and of the alveolar nerve. Since normal transaxial. CT images do not meet these requirements special secondary reconstructions adapted to jaw anatomy are necessary: -panoramic secondary cuts The cut line follows jaw curvature and represents a similar view as orthopantomographic pictures. (see Fig.1) -oblique secondary cuts That are plane cuts perpendicular to the panoramic cut line. (see Fig.2)

In biology and medicine there are many problems concrninq in vestigations of the human skeleton Beside the long bones the skull is the most important part of the skeleton for m.surement and iden tifiction The fc:e part9 the brain part and the mandibular part of the skul I are qenetical ly determined sectors with high percritage of individual characteristics

There are different imaging techniques in nuclear medicine for cardio- logic diagnostics, the specific statement of each determines the position in the indication spectrum. Moreover the burden for the patient and the methodic expence affect their use. Properly speaking medical imaging of the heart is a 4'dimensional problem three spatial dimensions and time as the fourth.

Results of routinely used frank leads do not relate to localization of the point of origin of late potentials (LP). In 13 pts. with anterior myocardial infarction (AMI) 12 with posterior rnyocardial infarction (PMI) und 6 with arrhythrnogenic right ventricular dysplasia (ARVD) high amplified signal averaged mapping with at least 16 leads has been performed. The maxima for LP duration voltage and calculated indica tors Ri and R2 are relatively homogeneously localized between left parasternal (PSL) and rnedioclavicular (MCL) line and the 4. - 6 rib. In PMI the maxima are less homogeneous but in the same area and in ARVD to the right of sagittal line. The maxima correlate to localiza tion of kinetc disturbances in MI. This technique does not provide exact localization of LP origin but helps planning endornap and has higher accuracy for risk prediction.

Various biomedical bbjects are 3D curveshaped. The paper discusses various methods a-F recovering the shape a-F such objects from multiframes under various geometrical and physical settings. 1.

The distinction between different types of giant-cell containing lesions often represents a challenge for pathologists since wide ranges and overlappings of their morphological features are common. But despite of their similarities the treatment is very different and depends on an accurate morpological diagnostic. Our present study was undertaken to investigate the nuclear DNA content (NDC) of giant-cell containing lesions and its value for the diagnostic. Furthermore we studied the significance of NDC to predict the biological behavior of giant-cell tumors. 12 aneurysmal bone cysts 25 central osteosarcomas (biopsies) and 16 giant-cell tumors of the Hamburger Knochentumorregister were included in this study. In all cases fresh material was available. 20 imprintcytologies from each specimen of each case were prepared . 10 were used for Feulgen-staining to perform quantitative DNA-measurements. 10 were used for Pappenheim-staining for morphological examination. Conventionally prepared sections of each specimen of each patient radiological findings and clinical data served as control-system. The recently developed DNA measuring program CESAR based on image analysis (IBAS I Kontron Germany) was used for this study. The obtained DNA distributions were classified according to a new developed DNAgrading system for imprintcytologies (s. table 1). All investigated aneurysmal bone cysts were 0 I (euploid). 23 central osteosarcomas were G III (aneuploid). 2 central osteosarcomas (predominantly cartilage producing) were G II (only a small number of hypertetraploid cells). 9 giant-cell tumors were G II. 7 Giant

Shape matching between a predefined shape model and an raw image is a key problem in image recognition, which can arise in character recognition, industrial applications, and medical applications. Medical images pose particular difficulties in that the image is noisy, the background is complex, and the shape is flexible. Dynamic programming can be applied to shape matching where the shape model is expressed by a sequence of line segments and the shape is varied by independently changing the length of the line segments. The generated shapes are applied to the raw image at every location. For each shape and location, the average angular similarity between the model segment and the output of an edge-operator applied along the line segment is computed. The optimal criterion is evaluated as the maximum of the similarity function and so forms a natural criterion to evaluate. However, there are a huge number of computations if we calculate by an exhaustive method. In order to decrease the computation while preserving the optimal solution, Dynamic Programming (DP) is introduced. The DP matching method is successfully applied to medical images of microscopic kidney tissue and ultrasonic image sequence showing heart action.

This review gives an inventory of methodologies used in three-dimensional imaging and visualization in biomedical sciences. It mainly addresses multislice data acquisition in microscopy and clinical scanners. Subjects treated for visual ization of volume data sets are three-dimensional reconstruction representation editing animation techniques and special hardware solutions. 1

Conventional image acquisition is based on fixed sampling standards. Little care is taken on special requirements regarding spatial and dynamicresolution as well as temporal behaviour ofthe objects. Newdevelopments in digital acquisition systems and sensor technologies offer the chance for dynamic adaptation of acquisition parameters to the object features. A loop for acquisition and parameter adaptation is shown. By that way we optimize not only image processing algorithms but also transmission and archiving solutions. The idea of adaptive acquisition is derived from the best image recognition system - thehuman eye and brain. 1.

This paper describes the analysis of human retinal blood vessels in terms of their fractal dimension ( D I . 7 ) . The fractal analysis of fundus images has applications to studies on the development of the retina and to diagnostic retinal imaging. Since the value of the fractal dimension for the retinal vessels is similar to that found for a diffusion limited growth process the results supports the hypothesis that diffusion is the fundamental process in the formation of human retinal vessel patterns. The fractal classification methodology permits the rapid screening of clinical furidus images and may have diagnostic potential. 1.

A newly developed imaging principle two dimensional microscanning with Piezo-controlled Aperture Displacement (PAD) allows for high image resolutions. The advantages of line scanners (high resolution) are combined with those of CCD area sensors (high light sensitivity geometrical accuracy and stability easy focussing illumination control and selection of field of view by means of TV real-time imaging). A custom designed sensor optimized for small sensor element apertures and color fidelity eliminates the need for color filter revolvers or mechanical shutters and guarantees good color convergence. By altering the computer controlled microscan patterns spatial and temporal resolution become interchangeable their product being a constant. The highest temporal resolution is TV real-time (50 fields/sec) the highest spatial resolution is 2994 x 2320 picture elements (Pels) for each of the three color channels (28 MBytes of raw image data in 8 see). Thus for the first time it becomes possible to take 35mm slide quality still color images of natural 3D scenes by purely electronic means. Nearly " square" Pels as well as hexagonal sampling schemes are possible. Excellent geometrical accuracy and low noise is guaranteed by sensor element (Sel) synchronous analog to digital conversion within the camera head. The cameras principle of operation and the procedure to calibrate the two-dimensional piezo-mechanical motion with an accuracy of better than O. 2. tm RMSE in image space is explained. The remaining positioning inaccuracy may be further

A digital video image analysis system is presented which consists of a personal computer equipped with a real-time video digitizer and a graphic tablet controlled by a modular set of programs aimed at performing a number of real-time oriented measuring tasks in microcirculatory research. Such tasks comprize the continuous recording of vessel diameters flow velocities or light intensity profiles from video recordings obtained during intravital microscopy of the terminal vascular bed either in research animals or in human beings. Two outstanding features of the presented systems are (A) the spatial correlation module for velocity measurement and (B) the automatic background movement correction. A: The spatial correlation velocity measurement module combined with an asymmetric illumination or gating process for image generation allows measurement of flow velocities from video microscopic images up to 20 mm/sec. This is about 10 to 20 times faster than the maximum. velocities which can be measured using conventional video based techniques. B: The automatic background movement correction is designed to track translational movements of background image structures in a reference window in real time (with respect to the video system). The translational vector of the image background is then used to adjust the position of the individual measuring lines or windows used in the different application modules to their original position relative to the tissue structures which are investigated. Such an automatic real-time tracking system isvery often a

In this study we describe three different approaches for the 3-D reconstruction of the spatial arrangement of iniranuclear chromatin. Using a Zeiss confocal laser scanning microscope (CLSM) image acquisition of optical sections was achieved in the fluorescence mode either using isolated nuclei stained with ethidiumbromide (EthBr) or on Feulgen stained tissue whole mounts. Different methods were applied for 3-D reconstruction: (i) contours of interesting structures were outlined by interactive cursor movement on a digitizer tablet (ii) digitized optical sections were transformed into image stacks by a software implemented on the microscope system and finally processed for a 3-D display and (iii) a ray-tracing method was used to provide a 3-D display of reconstructed surfaces from serial CLSM images after extensive image preprocessing. The characteristics of the different methods are discussed with respect to the biological system used.

Experimental and clinical studies have proved the practicability and feasibility of employing three-dimensional ultrasonic diagnostic techniques in a clinical context. Three-dimensional ultrasonic imaging originated from the idea that by rotating the scan plane around a fixed centre, a defined series of slices ( sectional images ) could be obtained that could be reconstructed to produce a three-dimensional display. Although three-dimensional reconstruction of parallel slices is the simplest solution - nuclear spin tomography and computerized tomography employ this technique - it is very difficult, on account of the unevenness of the surface of the body, to obtain parallel ultrasonic slices by means of ultrasound in clinical applications. Since a three- dimensional reconstruction is based on a coordinated sequence of slices, the simplest method of obtaining the three-dimensional ultrasonic display would seem to be by means of the above- mentioned rotated scan plane. This, however, requires a new scanning head to be designed in order to obtain such a slice sequence. TFte spatial relationship between the individual slices has to be fed into a hooked-up computer and suitable software for the contouring and spatial array of these slices needs to be developed.

Laser scanning tomography is a technique for obtaining three-dimensional morphometric measurements in vivo. Beside " volumetric" parameters for optic disk analysis eg excavation volume maximal excavation depth or third moment of frequency distribution of excavation depth readings the evaluation of a new parameter the contour-line modulation is of special interest. The contour-line defines the disk border. In laser scanning tomography each pixel of the contourline is defined three-dimensionally. The contour-line modulation is the difference in height readings between the mean height position of the contour-line in the segment under investigation and the mean height position of the contour-line of an optic disk. We analysed a series of 27 control group eyes of 27 patients (C no defects in computerized static 30 degree visual field test octopus program 32 or 38) and 38 glaucoma eyes of 38 patients (G at least three points with a relative scotoma of -1 0dB loss or greater). Within the temporal 1 800 of the optic nerve head the contour-line modulation values were +79 m for the temporal superior octant (G) and +40 m for the temporal inferior octant (G) compared to the temporal quadrant and +1 71 pm for the temporal superior octant (C) and +1 34 tm for the temporal inferior octant (C) respectively . These data suggest that contour-line modulation values may be a quantitative structural parameter for nerve fiber bundle defects. Further studies

Local retinal sensitivities are measured by projecting stimuli directly onto the retina. In this way stimuli can be positioned more accurately than with previous technologies by using a Scanning Laser Ophthalmoscope (SLO) with a computer and a modulator. These devices constitute a system which has to be investigated and calibrated to ensure proper accuracy in measurements. After developing a prototype computer program for manual static Fundus Perimetry system characteristics could be calibrated and first measurements were obtained. 1 .

A new optical method for measuring eye movements in three dimensions is presented. It is based on imaging the ocular fundus with a scanning laser ophthalmoscope (SLO) and determining the rotational state of the eye from the positions of retinal features in subsequent images. The hardware required for image acquisition as well as algorithms for image correction fundus tracking and 3-D data processing are described. The method operates linearly within deg about the primary eye position with an angular resolution of 0. 1 deg and a sampling rate of 50 Hz. The kinematic parameters of an eye rotation are expressed as (1) rotation vector (2) Euler angles and (3) quaternions. Additionally the SLO allows precise retinotopic visual stimulation (via modulating the laser intensity) thereby rendering the evaluation of the correlation between stimulus movements and eye movements feasible. Examples of eye movement recordings are presented. 1.

Confocal microscopy of ocular tissue provides two advantages over traditional imaging techniques: increased range and transverse resolution and increased contrast. The semitransparent cornea and ocular lens in the living eye can be optically sectioned and observed by reflected light confocal microscopy. Within the cornea we observed various cell components nerve fibers nerve cell bodies and fibrous networks. The confocal microscopic images from the in-situ ocular lens show the lens capsule the lens epithelium and the individual lens fibrils. All of the reflected light confocal microscopic images have high contrast and high resolution. Some of the applications of confocal imaging in ophthalmology include: diagnostics of the cornea and the ocular lens examination prior to and after refractive surgery examination of intraocular lenses (IOL) and examination of eye bank material. Other ophthalmic uses of confocal imaging include: studies of wound healing therapeutics and the effects of contact lenses on the cornea. The proposed features of a clinical confocal microscope are reviewed. 2.

Expert systems are exciting for primarily two reasons: first they are generally useful, practical programs that fulfil an actual need; and second they are available. These are the reasons that they constitute most of AI's practical and commercial success /19/. At their best, expert systems may be as helpful as having an expert consultant right there /2/. Expert systems are computer software that can function as a consultant, providing guidance advice and assistance in decision making, handling automation of human's ability in sharply defined and delineated regions. In this sense it is possible to replace a human expert by an expert system /3/. Because human expertise is rare and expansive expert systems are needed /18/.