Aspects of the organization of neurons and dendritic bundles in primary somatosensory cortex of the rat.

In order to analyze some aspects of the spatial organization in the primary somatosensory cortex of the rat, we have reconstructed the positions of bundles of apical dendrites and neurons in a cortical prisms measuring 0.5 mm x 0.4 mm x cortical thickness, with special reference to a hypothetical columnar organization. Complete series of semithin (0.65 microm) sections were cut, tangentially from the pial surface down to the white matter, stained and digitizalized into a computer and represented as a stack of 2D images. The mean neuron density (N(V)-value) was (60 x 10(3) +/- 15 x 10(3)) neurons/mm3. The mean number of neurons beneath 1 mm2 of cortical surface (NC-value) was (113 x 10(3) +/- 8 x 10(3)) neurons/mm2. Well-defined bundles of apical dendrites emanating from layer V pyramidal cells were observed. The bundles consisted of 3-12 (mean 5 +/- 2) dendrites. The dendrites within a bundle converged while ascending towards the pial surface and reached a maximal close packing in layer IV. Superficially, the packing density decreased again. The mutual positions of the dendrites within the bundles shifted only slightly along their course towards the pial surface. The occurrence of bundles in tangential sections through layer IV was about 190 bundles/mm2 and the average number of neurons per bundle was estimated at approximately 600. However, when calculating Voronoi-diagrams, the number of neurons, which with this mathematical technique, is ascribed to each of the reconstructed dendritic bundles, varied between 200 and 1000. The possibility that the dendritic bundles are centers in cortical cell modules is discussed.

3D visualization and stereographic techniques for medical research and education.

While computers have been able to work with true 3D models for a long time, the same does not apply to the users in common. Over the years, a number of 3D visualization techniques have been developed to enable a scientist or a student, to see not only a flat representation of an object, but also an approximation of its Z-axis. In addition to the traditional flat image representation of a 3D object, at least four established methodologies exist: Stereo pairs. Using image analysis tools or 3D software, a set of images can be made, each representing the left and the right eye view of an object. Placed next to each other and viewed through a separator, the three dimensionality of an object can be perceived. While this is usually done on still images, tests at Mednet have shown this to work with interactively animated models as well. However, this technique requires some training and experience. Pseudo3D, such as VRML or QuickTime VR, where the interactive manipulation of a 3D model lets the user achieve a sense of the model's true proportions. While this technique works reasonably well, it is not a "true" stereographic visualization technique. Red/Green separation, i.e. "the traditional 3D image" where a red and a green representation of a model is superimposed at an angle corresponding to the viewing angle of the eyes and by using a similar set of eyeglasses, a person can create a mental 3D image. The end result does produce a sense of 3D but the effect is difficult to maintain. Alternating left/right eye systems. These systems (typified by the StereoGraphics CrystalEyes system) let the computer display a "left eye" image followed by a "right eye" image while simultaneously triggering the eyepiece to alternatively make one eye "blind". When run at 60 Hz or higher, the brain will fuse the left/right images together and the user will effectively see a 3D object. Depending on configurations, the alternating systems run at between 50 and 60 Hz, thereby creating a flickering effect, which is strenuous for prolonged use. However, all of the above have one or more drawbacks such as high costs, poor quality and localized use. A fifth system, recently released by Barco Systems, modifies the CrystalEyes system by projecting two superimposed images, using polarized light, with the wave plane of the left image at right angle to that of the right image. By using polarized glasses, each eye will see the appropriate image and true stereographic vision is achieved. While the system requires very expensive hardware, it solves some of the more important problems mentioned above, such as the capacity to use higher frame rates and the ability to display images to a large audience. Mednet has instigated a research project which uses reconstructed models from the central nervous system (human brain and basal ganglia, cortex, dendrites and dendritic spines) and peripheral nervous system (nodes of Ranvier and axoplasmic areas). The aim is to modify the models to fit the different visualization techniques mentioned above and compare a group of users perceived degree of 3D for each technique.

Virtual reality on the web: the potentials of different methodologies and visualization techniques for scientific research and medical education.

Academic and medical imaging are increasingly using computer based 3D reconstruction and/or visualization. Three-dimensional interactive models play a major role in areas such as preclinical medical education, clinical visualization and medical research. While 3D is comparably easy to do on a high end workstations, distribution and use of interactive 3D graphics necessitate the use of personal computers and the web. Several new techniques have been demonstrated providing interactive 3D via a web browser thereby allowing a limited version of VR to be experienced by a larger majority of students, medical practitioners and researchers. These techniques include QuickTimeVR2 (QTVR), VRML2, QuickDraw3D, OpenGL and Java3D. In order to test the usability of the different techniques, Mednet have initiated a number of projects designed to evaluate the potentials of 3D techniques for scientific reporting, clinical visualization and medical education. These include datasets created by manual tracing followed by triangulation, smoothing and 3D visualization, MRI or high-resolution laserscanning. Preliminary results indicate that both VRML and QTVR fulfills most of the requirements of web based, interactive 3D visualization, whereas QuickDraw3D is too limited. Presently, the JAVA 3D has not yet reached a level where in depth testing is possible. The use of high-resolution laserscanning is an important addition to 3D digitization.

3D-brain 2.0--narrowing the gap between personal computers and high end workstations.

UNLABELLED: Recent advances in personal computer hardware and software have pushed the graphic capacity of these easier to use and, more importantly, cheaper computers to a level approximating the current standard of high end workstations. The interactivity and graphic complexity of a modern PC is rapidly approaching the current standard on Silicon Graphics (although with respect to texture mapping, the SGI is still ahead of the PCs). The modern medical student laboring under increasingly higher demands with respect to versatility, not only in basic science and traditional medical knowledge, is also faced with the requirement to learn and understand modern scientific visualization and analytical instruments. Furthermore, basic knowledge of information technology and computer literacy is expected of the next generation medical professionals. These demands forces medical schools to increasingly invest in computers and information technology for educational purposes. Due to common class sizes, these computers are most commonly Windows PCs or Apple Macintoshes. For distance education, telematics or studies at home, personal computer versions of the workstation graphics are a necessity. 3D-Brain 2.0 is an educational software package intended to run on basic personal computers and utilizing modern software technologies such as QuickTime VR 2.0 and VRML 2.0, to provide the students with insight into modern clinical and scientific visualization, focusing on the anatomy and functionality of the human brain. The aim of this paper to test the validity and usefulness of these new visualization techniques. METHODS: 3D-Brain is based on human brains sliced in 1 mm sections (NB. NOT based on NLMs Visual Human). Each slice was photographed, digitized, optimized and aligned using proprietary software. The datasets were then created by manual tracing followed by triangulation, smoothing and 3D visualization using Silicon Graphics computers. For the QuickTime VR project, 684 images with a 10 degrees angle were generated for each scene and ported to an Apple Macintosh computer for further manipulation. VRML code was generated directly from the original dataset. All interactivity was programmed on a Macintosh and subsequently ported to the Windows95 PC platform. The minimum requirements to run the software are either a PowerPC based Macintosh computer or a Pentium based Windows 95 computer with 16 Mb, 16 bit display and a 4 speed CD-ROM. RESULTS AND DISCUSSION: 3D-Brain 2.0 provides medical students at Goteborg University the means to complement traditional teaching using visualization techniques and three-dimensional models. These techniques also serve as an insight into the different clinical means of visualization the student will encounter throughout his/her continued education and professional career. For educational purposes, it has been established that among the tested new visualization techniques, CD-ROM based software utilizing QTVR is still the best methodology to use for pedagogical software. VRML shows promise in porting these software packages to the Web while Open Inventor is the preferred format for research purposes.

The existence of a layer IV in the rat motor cortex.

We have reconstructed the laminar pattern of rat primary motor cortex (Fr1) using a computerized analysis system based on the so-called 'optical dissector'. Data were visualized on a graphics terminal. In contrast to current views, which state that there is no prominent layer IV in the motor cortex of the rat, our method of analysis revealed a genuine layer IV consisting of densely packed small neurons.

Aspects of the quantitative analysis of neurons in the cerebral cortex.

We address three problems concerning the quantitative analysis of nerve cell distribution in the cerebral cortex: (i) preparatory tissue deformation (shrinkage); (ii) difficulties in differentiating between small neurons and astroglia; and (iii) the bias introduced by the counting method. We found that staining with Richardson's solution led to no shrinkage in Vibratome-cut sections of aldehyde-fixed rat brains, but did result in staining of the neurons and left the glial cells unstained. This was in striking contrast to Nissl staining which introduced a linear shrinkage of 20-30% and stained all kinds of cortical cells indiscriminately. A computer-based unbiased counting method was implemented by taking advantage of the stereological procedure referred to as the 'optical disector' (Gundersen, H.J.G. (1986) Stereology of arbitrary particles, J. Microsc., 143: 3-45).

Heterogeneity in the columnar number of neurons in different neocortical areas in the rat.

We have investigated the number of neurons in three neocortical areas of the rat brain. Our results challenge the uniformity concept proposed by Rockel et al. [Brain, 103 (1980) 221-244]. Area Fr1, HL and Oc2 (primary motor, primary somatosensory and secondary visual cortex) from Sprague-Dawley rats were examined. The brains were glutaraldehyde fixed, sectioned in 50 mu m thick sagittal slices and stained in Richardson's solution. The counting was carried out using a computerized system based on the optical disector. The cortical thickness was measured to be 1.9 mm, 1.9 mm, and 1.4 mm in area Fr1, HL, and Oc2, respectively. The number of neurons under 1 mm2 cortical surface was calculated to be 91 100 in Fr1, 133 500 in HL and 106 100 in Oc2. The number of neurons in a volume of tissue 30 x 25 mu m through the depth of the cortex was calculated to be 68 in Fr1, 100 in HL and 80 in Oc2. The density of neurons was calculated to be 48 500 neurons/mm3 in Fr1, 69 400 neurons/mm3 in HL and 76,900 neurons/mm3 in Oc2. There were significant (P < 0.01) differences between all areas regarding both the number of neurons under a certain area of surface as well as the neuron density. The results indicate that there is no basic uniformity in the number of neurons under a certain area of cortical surface.

A method for 3D reconstruction of neuronal processes using semithin serial sections displayed as a cinematographic sequence.

In order to study the organization and distribution of dendrites and axons in the cerebral cortex, we have developed a computer-assisted method for 3D reconstruction of neuronal processes based on serial light microscopic images displayed as a continuous sequence. A series of tangential sections (0.65 micron thick) through rat parietal cortex was aligned, digitized into the computer and then used to build a sequence (stack) of images which was stored to a digital real-time video disk. Apical dendrites located in dendritic bundles in laminae III and IV were traced through the sequence. Two tracing modes were tested: (1) cinematographic mode, in which the image stack was displayed continuously and automatically by the computer at various preset speeds (max. speed: 25 images/s) and (2) stepping mode, in which the interval between each image was varied manually according to the choice of the operator. Coordinates were stored in a database and used to build a 3D reconstruction where apical dendrites were displayed as wires or tubes. Tracing in cinematographic mode was about 3 times faster than tracing in stepping mode. We believe that the former mode exploits the built in 'filtering' capacity of the visual system to perform temporal averaging.

Automated correction of linear deformation due to sectioning in serial micrographs.

This paper describes an objective and automatic method for detection and correction of sectioning deformations in digitized micrographs, as well as an evaluation of the method applied to light and electron microscopic images of semi-thin and ultra-thin serial sections from brain cortex. The detection is based on matching of image subregions and the deformation model is bi-linear, i.e. two first-order polynomials are used for modelling compression/expansion in perpendicular directions. The procedure is applicable to prealigned serial two-dimensional sections and is primarily aimed at three-dimensional reconstruction of tissue samples consisting of a large number of cells with random distribution and morphology.

Model based dynamic 3D reconstruction and display of the left ventricle from 2D cross-sectional echocardiograms.

In this paper we describe a technique for dynamic three-dimensional (3D) reconstruction of the left ventricle, using boundaries from multiple two-dimensional (2D) echocardiographic views. We use a geometric model of the left ventricle and anatomical landmarks to relate the recorded views to positions within the model. The reconstruction is step-wise refined by replacing model data with recorded contour data.

Computer-assisted 3D analysis of cell distributions in the normal and epileptic cerebral cortex: description of a methodology in progress.

This paper describes software routines that (a) visualizes a stack of several thousands of aligned sequential photographic two-dimensional (2D) images stored in an image processing system; (b) creates a data base containing information about objects identified sequentially from the 2D images; (c) transfers the data base to a graphical terminal; (d) reconstructs a three-dimensional (3D) object space; and (e) supports on-line interaction between the image processing system and the graphical terminal. As an application example, the cell content of a prism of motor cerebral cortex of the cat is reconstructed. Preliminary results from reconstructing human epileptic temporal cortex (cortical microdysgenesia) are also reported.

3D reconstruction of biological objects from sequential image planes--applied on cerebral cortex from cat.

A prism of cat cerebral cortex was reconstructed with a method for three-dimensional (3D) representation of biological objects. A series of 918 semithin sections were digitized into an image analysis system. The images were aligned and analyzed, and a data base with the coordinates and a classification of the cells was created. The data base (i.e., the cortical prism) was visualized in a 3D graphic terminal, and parameters such as columnar and lamellar organization, clustering, and cell density were analyzed. A neuronal perikaryon and its neurites was reconstructed and shown together with the cortical prism.