The results of our previous studies revealed that the specimen heating holder with the heating elements of spiral shaped fine metal wires of high melting point enable us to observe high resolution transmission electron microscopy(TEM) images at elevated temperatures.

In fact, the holder was applied for high resolution TEM study of a formation of SiC crystal at 1500°C and a surface reconstruction of Au deposited Si particle at 1000°C successfully. However, because the heating holder was single tilt type, there was a certain limitation in its application.

In this paper, development of a double tilt specimen heating holder with a heating element of spiral shaped fine metal wire and its application for the study of microstructural changes of Si3N4 during sintering at very high temperature.

Photograph of the newly developed double tilt specimen heating holder is shown in Fig. 1. The heating element is mounted on the electrically isolated tilting frame of the holder and the heating current is supplied via tilting rod which is also electrically isolated from other parts of the holder.

The increasingly wide use of digital computers, the world-wide-web and electronic publishing has had a fundamental impact on the way scientists conduct research in every discipline of science. Electron microscopy is no exception. A considerable amount of effort has been devoted to the development of digital imaging acquisition systems for transmission electron microscopy (TEM). Digital image acquisition systems for TEM, including complete systems, have been produced by several companies, including: Advanced Microscopy Techniques (Rowley, MA), JEOL (Peabody, MA), Gatan (Warrendale, PA), Princeton Instruments (Trenton, NJ) and Tietz-Video (Herbststrasse, Gauting, Germany). While most systems are CCD-based, JEOL has also offered a system which is based on the Imaging Plate technology.

The Imaging Plate has the same size as, and is compatible with the camera system for, the 8.09 cm × 99.6 cm electron microscope film. As with film, a latent image is formed on the plate when exposed to electrons. A stack of exposed imaging plates are then taken out of the microscope and scanned by a laser beam in a readout device which converts the latent image to a digital form.

An event-driven 2D pixel array detector designed for protein crystallography is being tested to determine its suitability as a direct electron detector for digital imaging in electron microscopy. The detector prototype consists of an 8×8 pixel array with an 150 μm pitch. The device configuration is illustrated in Fig. 1. The detector module consists of a reverse-biased 300 μm thick monolithic silicon diode array which is bump-bonded to an Application Specific Integrated Circuit (ASIC) which instruments each diode in the array. Each pixel cell of the ASIC forms a pixel processor which consists of an amplifier, discriminator, prescalar and readout circuit.

When an incident electron enters the detector, a group of electron-hole pairs is created in the silicon. The detector bias of the diode array is such that holes are collected at the input to the amplifier where the signal is then amplified and shaped. When the output signal pulse height is greater than the discriminator threshold, the event is then registered in the prescalar, causing one count to be accumulated.

The performance of slow scan CCD (SSC) camera [1,2] and imaging plate (IP) [3] for digital recording is limited by resolution and noise. The SSC camera and IP are two digital detectors currently available for transmission electron microscope (TEM). Imaging plates are re-usable flexible sheets, which are used in the standard cassettes. They are readout digitally with a 25 μm pixel size. Both detectors are linear and have large dynamic range. The SSC is a fixed accessory of TEM, images can be acquired, processed and viewed immediately by the microscope operator. The IP fits into a regular film cassette and is used like film except for the processing method. The image plate records electron image by storing electron energy in the potential well of defect states in a photo-stimulable phosphor. The stored image is read out by scanning a laser probe and detecting the stimulated luminescence in a reader. The SSC uses scintillator to convert electrons to photons and detects photons with CCD through optical couplings. The IP has 3000X3760 pixels. The available SSC for electron microscopy ranges from 512×512 to 2048×2048 pixels.

All the lenses, deflectors and stigmators of contemporary electron microscopes are controlled digitally by an internal computer. Control through RS232 serial interface by an external computer has also become a standard feature. This external control has made so-called computer assisted electron microscopy (CAEM) possible and practical. We are developing a CAEM system with two objectives: (1) to free inexperienced microscopists from technical details of operating an electron microscope, especially transmission electron microscopes (TEM); (2) to assist experienced microscopists to operate their microscopes with higher accuracy and efficiency. The features include automated and/or assisted standard operations in focusing, stigmating, and aligning the microscope, and also sophisticated tuning that requires the evaluation of subtle changes in image features such as aligning the incident electron beam direction in the presence of 3-fold astigmatism in objective lens. CAEM can further assist operators in selecting areas or objects and taking images/diffraction/energy spectrum with all the parameters well controlled and catalogued together, thus not only enabling ease-of-use and high accuracy in operation but also yielding more information on the specimen.

The Lehigh University is presently offering direct access to top of the line electron microscopes at ORNL via the Internet for one graduate and one undergraduate course (Advanced TEM, Electron Microscopy & Microanalysis). The students at Lehigh have the opportunity to themselves control the Hitach HF-2000 FEG TEM. The controls presently include change of magnification and focus, moving the sample and correction for astigmatism. In this way, the students can collect high resolution lattice images of various materials and study e.g., the effects of changing focus on lattice images (contrast reversal). While the students at Lehigh can sit in the class room and run the instrument in broad daylight and communicate freely without disturbing the instrument by phonics, the instrument itself is protected from misuse through several layers of hardware and software.

This new way of teaching EM courses has become possible through recent advances in computer hardware (live-time Fourier transforms (FFTs)) and software. A commercial software package TimbuktuPro provides remote access to the computer at the HF-2000 at ORNL which controls the instrument as well as the slow-scan CCD-cameras (Gatan, model 794) attached to the instrument.

Digital imaging has provided a number of possibilities in optical microscopy which can not be achieved easily by traditional methods. These include the possibility of low-light imaging, ease of image manipulation (processing and analysis) and transmission via electronic communication channels. Recently, digital imaging has become the buzz-word in photography. Therefore, it is the intent of this article to evaluate the current state of digital imaging technologies for optical microscopy, both pro and con.

There are certain applications where digital imaging is the only feasible way to gather image data, examples include laser scanning confocal microscopy and low-light CCD microscopy. Based on the nature of image acquisition methods, one can classify digital imaging into two major domains: (1) sequential and (2) parallel data acquisition. The sequential data acquisition method is generally used in scanning devices such as the confocal laser scanning microscopes (CLSM). Photomultipliers and solid state devices are generally used as the photon detector(s) and the analog signals are subsequently digitized. On the other hand, the use of 2D array detectors such as CCDs falls into the parallel acquisition category. Although, the data from a CCD is read-out sequentially, all the pixels on a CCD chip acquire the image simultaneously. Depending on the dynamic range of the photon detector and needs of the application, the digitized image generally has a dynamic range of 8, 12 or 16 bits.

The principle of spatial mapping is familar to microscopists. It is, for example, employed in conjunction with x-ray fluorescence spectrometry to draw up maps showing the distribution of elements in a materials surface. Although widely exploited in scanning electron microscopy, the technique of spatial mapping is rarely applied to other types of images. With the recent development of automated methods for recognizing and extracting details from digital images, however, it has become feasible to extract patterns from the transmission electron microscope image and overlay the original image with representations from the data. Pattern extraction, executed in a data mining scheme, is an area of common interest to researchers in machine discovery, data visualization, and high performance computing. Since the resolution of an electron microscopic image, if completely conserved in the conversion into digital format, can range into many millions of pixels, microscopy is one of the most promising areas for data mining applications.

This capability is particularly well-suited to investigating the problem of chromatin higher order structure. The highest order of packing that is readily seen in transmission electron micrographs of chromatin is the 30 nm fiber, the approximate size of which has been confirmed by x-ray diffraction studies.

Digital image acquisition in the SEM is poised to take over conventional methods of image recording. Because a majority of researchers use Polaroid film which only records about 1,000 lines of information, the acquisition systems available today are capable of matching and possibly exceeding the resolution of this medium.

Several methods are available to accomplish the acquisition task, and they each have certain benefits and disadvantages. by far the simplest and least expensive is to digitize the video output from the SEM when available. Under some conditions the video output can be digitized directly. The advantage of this method is cost. A single frame video digitizer costs less than $200. The signal, however, is usually very noisy. Because the human eye “sees” at a slower rate than video, we normally are “seeing” a 10-20 frame average. When we choose one frame, the results are too noisy. Fortunately, for under $1,500, digitizers that average frames are available. With multiple frames, the results are often acceptable for some applications.

For the resolution estimation of the Imaging Plate (IP)-system, several methods were carried out. One of them is the measurement of the modulation transfer function (MTF) from the phase contrast transfer function taken with an amorphous Ge film specimen. One of another is the measurement of the point-spread function (PSF) from the auto-correlation of the Poison noise obtained by the IP-system.

In this experiment, the PSF of the IP-system were measured from the intensity distribution recorded with a 1.6 μm-diameter electron probe on the IP.

When recording the electron intensity with a small enough electron probe as compared with the pixel size of the IP (25 μm × 25 μm), a single pixel detects the electron signal in case of ideal IP-resolution which is limited only by the pixel size. However, practical IP-resolution is limited by the electron diffusion, x-ray emission and the laser beam diffusion in the IP, and so on. Therefore, it is possible to measure the PSF from the intensity distributions of the patterns recorded with a small enough electron probe.

Recently, new recording devices, such as the Imaging Plate and a slow scan CCD camera, have come into wide use for digitally recording and storing transmission electron microscope images. A digital image has advantages of transferring, sharing and storing data. Furthermore the improvement of world wide computer network systems, for example, Internet makes it easier to transfer and share digital data among researchers. We have studied the processing and a quantitative analysis of digital high-resolution transmission electron microscope (HREM) images by means of a standard EWS and an SX-3 super-computer on the intelligent computer network system in Tohoku University, which is called Super TAINS (Fig. 1). With the increase of digital HREM images, we are in need of tools to help us store and quickly search these data.

A web site called “EMILIA (Electron Microscope Image Library and Archive)” was established in October 1996 as a database of HREM images. EMILIA is implemented by a WWW server on Super TAINS and HTML documents. Figure 2 shows sample pages from EMILIA.

The new generation of cooled slow-scan CCD cameras (SSC) have demonstrated the superior properties to conventional photographic films in sensitivity, resolution, dynamic range, linearity, and on-line availability. Furthermore the digital imaging capability as offered by SSC cameras facilitates image processing and quantitative analysis. Other advantages of digital imaging include easy archiving, sending images to a remote site via high speed communication network, and carrying out telemicroscopy, etc. Today, many laboratories in electron microscopy have already abandoned the traditional “dark room” and switched to digital imaging. One of the obstacle in replacing photographic films with digital imaging is the fact that current SSC cameras (e.g. 1024 × 1024 pixels) do not have enough pixels as compared with the film. Therefore the resolution is often compromised with the field of view when using CCD cameras.

With computer control of electron microscopes and the use of CCD cameras, we have developed automated “montage” software that can automatically shift and acquire images from a microscope, and automatically perform pixel-to-pixel image alignment to form a montage. The size of the montage is defined by the number of images acquired in both horizontal and vertical directions.

Manual scanning electron microscope (SEM) analysis is historically considered to be slow and tedious resulting in a low volume of data. This is due in large part to the mechanics of moving stage locations and recording image and spectral data. Conversely, high volume data acquired using automated SEM analysis has been associated with the need for complex systems for data management and analysis. In addition, the proliferation of high volume digital microscopy and its attendant “ tonnage” of paper images has lead to the desire for a “green” (filmless and hardcopy-reduced) operation.

There are some classes of projects which are amenable to automated feature analysis - discrete features that are distinct from a background material. However, many projects require operator intervention in order to identify the region or points of interest. Yet, these projects may also require that large data sets be acquired and analyzed for statistical rigor.

Electron tomography is a powerful tool for elucidating the three-dimensional architecture of large biological complexes and subcellular organelles. Use of intermediate voltage electron microscopes extended the technique by providing the means to examine very large and non-symmetrical subcellular organelles, at resolutions beyond what would be possible using light microscopy. Recent studies using electron tomography on a variety cellular organelles and assemblies such as centrosomes (Moritz et al.,1995a,b), kinetochores (McEwen, 1993) and chromatin (Woodcock, 1994), have clearly demonstrated the power of this method for obtaining 3D structural information on non-symmetric cell components. When combined with biochemical and molecular observations, these 3D reconstructions have provided significant new insights into biological function.

Although the information that tomography provides is unique, its use as a general tool in the biological community has been limited due to the complexities involved in data collection and processing.

We are simultaneously trying to make this approach accessible through automation as well as trying to extend the resolution and accuracy of the reconstructions. Significant, has been the use of low-dose cryo-electron microscopic automated data collection methods.

Recording electron microscope images of frozen-hydrated samples under cryo-conditions is a difficult and challenging task. The inherently low sample contrast makes accurate focusing difficult. Furthermore most frozen-hydrated samples are very beam sensitive so that electron dose on the sample must be kept to the minimum. In recent years, the new generation of cooled slow-scan CCD (SSC) cameras have shown superior properties in sensitivity, resolution, dynamic range, linearity, and more importantly their ability to display images on-line. Computer control of electron microscopes via the standard RS232C serial interface has been greatly facilitated by the use of on-line CCD images. We have developed a software package that allows cryo-electron microscope images to be automatically acquired onto a CCD camera via computer control of the microscope. This fully automated feature allows quality cryo-electron microscope images to be recorded with increased efficiency and enhances the possibility of obtaining three-dimensional reconstructions from images.

Cryo-electron microscopy allows high resolution structural studies of macromolecules or macro-molecular complexes. As structural analyses extend to higher resolutions, several major compu-tational problems arise in analyzing cryo-electron micrographs. One is the acute sensitivity of the sample to radiation damage, requiring that images be acquired at a low electron dose with conse-quently low signal-to-noise ratio (SNR), especially at higher spatial frequencies. Secondly, as the size of each digitized image grows, the duration of the computational procedures lengthen consider-ably. A third problem is the complex distortion imposed upon the images by the contrast transfer function (CTF) of the electron microscope. We have addressed these issues in the context of solving the Hepatitis B Virus (HBV) capsid structure, and have succeeded in improving the resolution of our model from 17À[4] to 9À[1] (Fig.l). This is sufficient to define part of the molecular structure, including a 4-helix bundle at the dimer interface which constitutes the protruding ‘spike’ domains seen on the surface of the capsid, as well as other helices elsewhere in the molecule.

Acquisition of CCD image data from frozen-hydrated specimens provides several advantages compared with the use of photographic film. CCD cameras possess excellent linearity, sensitivity and dynamic range. The CCD also allows one to pre-check the quality of a cryo sample much easier than with a TV rate camera or on film. However, the limited field of view (IK2 pixels) of the CCD makes data collection difficult. Gatan, Inc. has produced a LowDose software package, in conjunction with their DigitalMicrograph™ environment, that automates this process and allows for efficient collection of larger amounts of data. This software, along with our own image analysis software, has facilitated the computation of a three-dimensional reconstruction (3D) from images of baculo-expressed, VP2 particles of Aleutian mink disease parvovirus (ADV).

ADV has a capsid (∼28.5 nm diam.) that contains two proteins (VP1-2) of which ∼90% is VP2. ADV infections lead to plasmacytosis in mink and are a significant problem worldwide.

Cryo-electron microscopy (cryo-EM) and image reconstruction techniques provide a powerful way to study three-dimensional structures of biological complexes. As the resolution of the resulting three-dimensional reconstructions is typically in the range of 10-35 À, information from other structural techniques is often invaluable in analyzing the biological structure. We present the use of crystallographic coordinates to determine the amplitude of motion of a viral receptor binding site and molecular models to assess the degree of quaternary conformational flexibility within a multimeric protein assembly.

In order to visualize the αv integrin receptor binding site on the surface of human adenovirus (150 MDa), we have reconstructed adenovirus in complex with a Fab fragment from a monoclonal antibody (DAV-1) targeted to this site. The peptide epitope of DAV-1 has been mapped by MALDI mass spectrometry to 9 residues within the penton base protein and includes the integrin binding Arg-Gly-Asp (RGD) residues.

Combining pathology, which reports the structural features of individual cells and their spatial organi-zation in a tissue specimen, with molecular biology techniques (immunocytochemistry and fluores-cence in situ hybridization, FISH) for detecting the distribution of specific molecular species in individual cells is a powerful approach for gaining insight into the underlying disease mechanisms of carcinogenesis. This approach requires analysis of thick (>20 ¼m) tissue sections in which cells are preserved intact within the context of their environment. 3D (confocal) microscope image acquisition followed by 3D image analysis (IA) for extracting quantitative information are then used for quantita-tive analysis of tissue features such as nuclear and/or cell volume, shape, total fluorescence or FISH signal number. An essential component of the IA is detection of the individual cells, or their nuclei since many molecular species of interest are localized within the nucleus (e.g. genetic aberrations). We present here a completely automatic, 3D algorithm for this task, which based on the Hough transform (HT).

The electron microscope is essential for resolving complex biological structures, such as mitochondria, which are too small to be viewed in detail with the light microscope. In contrast to a conventional instrument, the High-Voltage Electron Microscope (HVEM) located at the National Center for Microscopy and Imaging Research (NCMIR) can obtain images from relatively thick specimens that contain substantial three-dimensional structure. Though a single image acquired with the HVEM represents a projection through the specimen, tomographic methods can be applied to a set of images acquired from different orientations to derive a three-dimensional representation of its biological structure. Tomography requires extensive computation and considerable processing time on conventional workstations in order to reconstruct the typically large HVEM volumes from the tilt series.

In order to expedite tomographic processing, we have implemented both the commonly used singleaxis tilt, R-weighted backprojection algorithm and two iterative reconstruction methods, algebraic reconstruction (ART) and simultaneous iterative reconstruction (SIRT) on the massively parallel Intel Paragon at the San Diego Supercompter Center.