For 350 years, the optical microscope has had a powerful symbiotic relationship with biology. Until this century, optical microscopy was the only means of examining cellular structure; in return, biologists have contributed greatly to the evolution of microscope design and technique. Recent advances in the detection and processing of optical images, together with methods for labelling specific biological molecules, have brought about a resurgence in the application of optical microscopy to the biological sciences. One of the areas in which optical microscopy is breaking new ground is in elucidating the large scale organization of chromatin in chromosomes and cell nuclei. Nevertheless, imaging the contents of the cell nucleus is a difficult challenge for light microscopy, for two principal reasons. First, the dimensions of all but the largest nuclear structures (nucleoli, vacuoles) are close to or below the resolving power of far field optics. Second, the native optical contrast properties of many important chromatin structures (eg. chromosome domains, centromere regions) are very weak, or essentially zero. As an extreme example, individual genes probably have nothing to distinguish them other than their sequence of DNA bases, which cannot be directly visualized with any current form of microscopy. Similarly, the interphase nucleus shows no direct visible evidence of focal chromatin domains. Thus, imaging of such entities depends heavily on contrast enhancement methods. The most promising of these is labelling DNA in situ using sequence-specific probes that may be visualized using fluorescent dyes. We have applied this method to detecting individual genes in metaphase chromosomes and interphase nuclei, and to imaging a number of DNA-containing structures including chromosome domains, metaphase chromosomes and centromere regions. We have also demonstrated the applicability of in situ fluorescent labelling to detecting numerical and structural abnormalities both in condensed metaphase chromosomes and in interphase nuclei. The ability to image the loci of fluorescent-labelled gene probes hybridized to chromosomes and to interphase nuclei will play a major role in the mapping of the human genome. This presentation is an overview of our laboratory's efforts to use confocal imaging to address fundamental questions about the structure and organization of genes, chromosomes and cell nuclei, and to develop applications useful in clinical diagnosis of inherited diseases.

We have examined the three-dimensional arrangement of chromosomes in embryos of Drosophila melanogaster at the syncytial blastoderm stage using three-dimensional optical sectioning microscopy. High-resolution optical sectioning in fixed embryos, in conjunction with in situ hybridization, has revealed the location of specific chromosomal regions within diploid interphase nuclei, as well as the spatial arrangement of mitotic chromosomes. Time-lapse in vivo optical sectioning has revealed the dynamic behavior of chromosomes throughout the mitotic cycle. Combination of these observations has provided insights into the dynamic aspects of three-dimensional chromosome behavior.

Confocal laser scanning microscopy (CLSM) is a significant improvement over conventional epifluorescence microscopy for observing biological structures. In addition to the increase in resolution and reduction of stray light by CLSM, the serial optical sections of fluorescently-labelled structures produced by CLSM are suitable for computer rendering techniques to produce three dimensional (3D) images of biological structure in the light microscope. The collection and properties of data sets obtained by CLSM and their subsequent computer rendering are described and the biological application of the technology is discussed and illustrated by reconstructions of fluorescently-labelled nuclei and mitotic spindles.

Drosophila melanogaster has become one of the most extensively studied organisms because of its amenability to genetic analysis. Unfortunately, the biochemistry and cell biology ofDrosophila has lagged behind. To this end we have been microinjecting fluorescently labelled proteins into the living embryo and observing the behavior of these proteins to determine their role in the cell cycle and development. Imaging of these fluorescent probes is an extremely important element to this form of analysis. We have taken advantage of the sensitivity and well behaved characteristics of the charge coupled device (CCD) camera in conjunction with digital image enhancement schemes to produce highly accurate images of these fluorescent probes in vivo. One of our major goals is to produce a detailed map of cell fate so that we can understand how fate is determined and maintained. In order produce such a detailed map, protocols for following the movements and mitotic behavior of a large number of cells in three dimensions over relatively long periods of time were developed. We will present our results using fluorescently labelled histone proteins as a marker for nuclear location1. In addition, we will also present our initial results using a photoactivatable analog of fluorescein to mark single cells so that their long range fate can be unambiguously determined.

Studies using fluorescent probes to determine the distribution of physiologically important ions and molecules have been the focus of much interest. Some of these probes may be used to measure the actual concentration of chemical species within cells if measurements are made at two or more wavelengths. However, because of the low photon levels of fluorescent emission at the single cell level, current intensified cameras yield useful information at about 1 second time resolution. Many biologically relevant change; in ionic composition in cells take place much more rapidly than this, therefore requiring a faster imaging system to study these processes in a localized manner. A fast digital fluorescence microscope capable of obtaining dual wavelength images at 30 images per second is described. It can record up to six minutes of data while switching excitation wavelength between image frames.

A method has been developed that permits the resolution of light microscopy to be extended below the fundamental limit imposed by the wavelength of light. The technique is based on our demonstration that significant light intensities can be passed through well-defined apertures that can be as small as 1/10 the wavelength of visible radiation.

We present an image restoration method that is suitable for images with missing or truncated data. This method, L2 regularization with a non-negativity constraint, is applied to image restoration of 3D optically sectioned microscope images of fluorescently labelled cells. Its ability to use axially truncated data is useful when applying it to 3D images from a conventional wide field microscope and to operate effectively on a small number of optical sections. Its ability to correctly place light originating from outside the field of view by making use of the out of focus information allows us to subdivide a large image into smaller pieces without a substantial edge effect. Large 3D data sets can be restored on a modestly priced computer workstation.

Volumetric data (scalar data sampled on a three dimensional grid) is effectively imaged by methods of direct volume rendering. We describe the volume rendering program RMSVOLUME created specifically for the rendering of data from macromolecular systems. Applications of the program to the display of electron density from xray crystallography and electron microscopy are presented.

We present a Fourier transform (Fr) spectrometer that is suitable for real-time spectral analysis in fluorescence imaging and flow cytometry. The instrument consists of (i) a novel type of interferometer that can be modulated at frequencies of up to 100 kHz and has a high light throughput; and (ii) a dedicated, parallel array processor for the realtime computation of spectral parameters. The data acquisition array processor can be programmed by a host computer to perform any desired linear transform on the interferogram and can thus separate contributions from multiple fluorochromes with overlapping emission spectra. We describe optics configurations for use in both flow cytometry and fluorescence microscopy. The integration of a flow cytometer and a spectral imaging fluorescence microscope is discussed, and the concepts of direct and reversed "virtual sorting" are introduced.

The 60 Hz field rate of RS-170 video systems is inadequate to capture the microscopic motion of rapidly moving biological specimens such as contracting skeletal and cardiac muscle. The recent introduction of low cost, high resolution CCD video cameras has permitted us to record muscle contractile dynamics without resorting to complicated and expensive non-standard video equipment. To this end, we have modified a frame-resettable charge-coupled device (CCD) camera for 60/240 Hz video field operation. The increase in image acquisition speed is achieved at the expense of vertical field size, but with no loss of horizontal resolution. For 240 Hz operation, there are no changes to the fundamental RS-170 video standard except for reinitializing the camera's scanning at a time earlier than normal. Thus, all RS-170 video devices work with this camera in either mode of operation without modification. Images are then digitized and stored in a computer field by field with a frame grabber for later image processing, analysis and measurement. This simple approach permits the detection and evaluation of rapidly moving microscopic specimens with appropriate aspect ratios to be made with relatively low cost RS-170 video equipment.

A series of simulations have been performed in order to investigate the feasibility of constructing a time-of-flight breast imaging system capable of achieving clinically useful resolution using state-of-the-art picosecond laser pulse generation and detection technology. Simulated images are presented which demonstrate the likely performance of such a system, based on the attenuation and scattering characteristics of breast tissue measured in the laboratory.

Hadamard transformed FTIR spectrometry combines high spectral resolution of FTS with spatial multiplex advantage of HTS. A Mylar film located in zone M and a Polyethylene film in zone P as sample is placed on the frame. The two-dimensional IR spectral images of above sample were obtained with a 255- element mask configuration folded into a 17x15 array on FTS.

Quantitative descriptions of features offer an objective cell classification. In addition, quantitative analysis is ideally suited for automated screening of samples for diseased (transformed) cells. More recently, it has been shown that subtle changes of DNA distribution in the nucleus can be detected by quantitative analysis which escape human detection . A key issue in quantitative analysis is the quality of the digital image which must be captured for such analysis . This paper focuses on the affect of image quality on the very important features--texture features--which have been shown to be the most powerful classifiers in quantitative pathology . Image quality affects these features in a direct and indirect way. The latter is the result of an effect on texture features by the segmentation process which in turn strongly depends on image quality.

Eigenanalysis is a powerful mathematical technique for analyzing matrices of data. With a data matrix constructed from a digitized image, this technique can be used to extract the features of the image. As a data processing methodology in image processing, the eigenanalysis is principally used in two ways. The first is to treat a single image as a data matrix. The second is to construct a data matrix from multiple images. In both cases, the input information is separated into mutually orthogonal eigenvectors obtained from the correlation or covariance matrix. Since the resulting eigenvectors are orthogonal, the information in each vector is excluded from all other vectors. Alternatively, the singular value decomposition method can be used to represent the data matrix as sum of its outer products, thus avoiding the construction of a correlation/covariance matrix. Both procedures allow sorting of information according to its significance, because the most significant information is associated with highest eigenvalues and corresponding eigenvectors. Consequently, the original data can be reconstituted and clustered using only the significant information. The advantage of this processing is that the preparatory artifacts of the sample and noise in the image are removed from the data. For applications in biological microscopy, the ultimate objective is to relate various structural patterns in the cell, enhanced by eigenanalysis, to their biological function.

Confocal bioimaging of the fine structure of the living rabbit cornea with both reflected light and fluorescent light has been demonstrated with a laser scanning confocal imaging system. Kalman averaging was used to reduce the noise in the images. Superficial epithelial, basal epithelial cells, stromal keratocytes, and endothelial cells were imaged. These cells and their subcellular structures were imaged in the two modes for comparison. The superficial epithelial cells were imaged by their autofluorescence (488/520 nm). This fluorescence signal may be due to the mitochondrial flavoproteins and can be used as a noninvasive indicator of cellular oxidative function. Thiazole orange was used to stain cell nuclei for fluorescence imaging. DiOC6 was used to stain the endoplasmic reticulum for fluorescence imaging. Fluorescein- conjugated phalloidin was used to stain actin for fluorescence imaging.

A multicolor fluorescence imaging system for tissue diagnostics has been constructed. Examples given to illustrate the system performance are atherosclerotic plaque lesions from human artery samples and a malignant rat brain tumor model. The system simultaneously monitors fluorescence images at four different wavelengths, enabling spatial as well as spectral information to be extracted. By selection of band pass filters an artificial image of the lesion under study can be formed, based on an optimal contrast function of four fluorescence intensities. The image is constructed in a computer and displayed in false colors on a monitor. Pulsed excitation light from an N₂ laser provides the possibility of using a gated detector, and in this way suppress room light to an undetectable level. Images can thus be recorded under white light illumination during visual inspection, increasing the usefulness of the system.

The number of photons that can be obtained from a fluorescent chromophore increases with the incident light intensity and the duration of illumination. However, saturation of the absorption transition and photodestruction place natural limits on the ultimate signal-to-noise ratio that can be obtained. Equations have been derived to describe the fluorescence-to-background-noise ratio in the presence of saturating light intensities and photodestruction. The fluorescence lifetime, the extinction coefficient, and the photodestruction quantum yield are the key parameters that determine the optimum light intensity and exposure time. This theory indicates that the laser power should be selected to give a mean time between absorptions approximately equal to the fluorescence decay rate, and the transit time should be selected to be nearly equal to the photodestruction time. Using these optimum conditions we have performed experiments to detect individual molecules of phycoerythrin (PE). The photocount distribution function, the photocount autocorrelation function, and the concentration dependence from phycoerythrin clearly show that we are detecting bursts of fluorescence from individual fluorophores. A hard-wired version of this single-molecule detection system was used to measure the concentration of PE down to 1015 M. This single-molecule counter is three orders-of-magnitude more sensitive than conventional fluorescence detection systems. The approach presented here has also been used to optimize fluorescence-detected DNA sequencing gels. Using a confocal microscope configuration we have detected DNA sequencing gels at concentrations as low as i0 fluorescent DNA fragments per band.

This paper discusses research directions and results from a multidisciplinary effort to develop feature extraction tools for analysis of the changes in molecular distribution during cell movement. This work is part of a broader effort directed at developing hardware for a new digital imaging microscope, as well as image restoration algorithms which precede the extraction steps, making them simpler. New voxel based display techniques are also being developed for improved visualization of the two and three dimensional data sets. The most complete feature extraction algorithm developed so far analyzes a time sequence of two-dimensional phase contrast images of newt eosinophilic granulocytes (white blood cells). It tracks a moving cell and also identifies the lamellipods of the cell. This allows the extraction of quantitative information relating cell motility to lamellipod formation. The algorithm finds the cell by finding those pixels in the image which belong to the boundary of the cell . Potential boundary pixels are identified by locating intensity changes due to the phase contrast halo surrounding the cell. While most boundary based image segmentation algorthms form a closed boundary by moving from a starting boundary pixel along a path which locally or globally optimizes a cost function our algorithm does not trace a path from a starting point and does not minimize a cost function. Instead, we close the boundary by examining the geometrical and topological relationships among potential boundary pixels. Gaps in the boundary are closed by connecting gap points to the "closest" boundary point. "Close" is determined by a distance metric which combines Euclidean and other types of geometric information about the boundary pixels already found The position of the cell in the previous image is used both to constrain the location of the cell in the image being examined and to insure that the boundary eventually found is indeed closed.

3D biomedical data obtained through tomography provide exceptional views of the interior structure of biological material. While visualization is one of the primary purposes for obtaining these data, other more sophisticated uses are possible. These include simulation of physical processes, interactive surgical investigation, modeling of tissue interrelationships, and analysis of dynamics. Visualization is merely the first contribution of computers to the display and analysis of these image data. Techniques are being developed to facilitate such advanced uses of 3D and 4D tomographic data.

The mechanisms of reorientation of individual DNA molecules undergoing Pulsed Field Gel Electrophoresis (PFGE) have been studied using T2 DNA molecules labeled with acridine orange and visualized with a fluorescence microscope. It is shown that molecules undergoing PFGE and conventional electrophoresis often get trapped in hook conformations (narrow U-shapes) that play an important role in determining the mobility of the molecules. It is found that the mechanism of formation of hooks require the previous generation of a kink (in which parts of the molecule double up inside a pore). Computer simulation experiments are presented to clarify the role of hook and kink formation in the size-dependent separation observed in PFGE experiments.