"Teaching Neuroanatomy with NeuroDatabase."

S.L. Wertheim, Ph.D.
University of Massachusetts Medical School
P.O. Box 390915
Cambridge, MA 02139
M.K. Wolf,
S. Billings-Gagliardi
Department of Cell Biology
A. Sams
Medical Center Library, University of Massachusetts Medical School
55 Lake Ave. North
Worcester, MA 01655
R. Ribitzky
Information Resources Division
University of Massachusetts
R.L. Sidman
Division of Neurogenetics, N.E. Regional Primate Research Center, Harvard Medical School, Southborough, MA.

Abstract
NeuroDatabase is a software system for the management of neuroscience images, related graphics and text. It is distinguished from most other computer-based neuroscience tools by the presence of a flexible underlying database. We will present the latest teaching functionality developed for a first-year medical school course in neuroanatomy.

PAPER

NeuroDatabase is a software system for the management of neuroscience images, related graphics and text. It has applications in both research and teaching (3,4,5). It is distinguished from most other computer-based neuroscience tools by the presence of a flexible underlying database. In teaching with NeuroDatabase, faculty at each institution can customize the information and its presentation according to local needs. NeuroDatabase allows teachers to incorporate their own images, and to organize them into one or more series. Each image can have any number of graphic overlays which indicate the position or borders of neuronal structures. Each structure can have associated neuronal connections (afferents and efferents) and clinical/functional correlations. A glossary and object typing scheme form an underlying knowledge base. Students can review and assess their own knowledge of the names, appearance, connections and clinical/functional correlations of brain structures via self-tests. Student use can be tracked and quantified.

Design and Implementation

The design and implementation of NeuroDatabase have been described in detail previously (1,2). The software is divided into graphical interface and relational database components. The graphical interface runs on Apple Macintosh computers and was built with SuperCard (Allegiant Technologies). A Macintosh IIci or better with 16 MB RAM is recommended for adequate performance. The software can run on 13" displays using 8-bit graphics, however it is both visually and educationally superior if run on 17" or larger displays using 24-bit graphics. The relational component requires Oracle (Oracle Corp.) and can run on a very wide range of hardware. The relational component can be run locally, on the same Macintosh, or (especially for multi-user access) across a local area network. Image files are stored externally to the database and can reside on a file server.

Typically, images are digitized using either a film scanner (for 35mm slides) or a flatbed scanner (for work on paper). Film scanners generally give superior results, both in terms of detail and color fidelity. Most of our images were scanned at approximately 1500x1000 pixel resolution, 24 bits per pixel, using either the Barneyscan (CIS) or Nikon LS-3500 film scanners. We have also used the Nikon LS-3500 scanner to digitize brain tissue cross-sections directly by putting the glass slide into the film holder. The images were then scaled, rotated and color-corrected as necessary using Photoshop (Adobe). Photoshop was also used to convert all images from 24-bit to 8-bit color and to create an optimized 8-bit color palette for each.

Image Series

Since almost all teaching images belong to a series of related images, the primary function of NeuroDatabase is image series management. NeuroDatabase allows instructors to add their own image series (once in digital form as described above), and to have multiple images of the same subject at different resolutions. Once an image series is incorporated into NeuroDatabase, users need to get an overview of the series, move through the images and request a particular image for close inspection. There are two primary mechanisms for this. The first, the Series Browser, shows small versions (icons) of all the images in a series. The icons are displayed in numerical order, four at once. A scroll bar allows the user to move through the entire series (Figure 1A). Clicking on an icon brings a detailed version of the image into an Image window (Figure 1B).

The second mechanism, the Overview window is shown in Figure 1C. The Overview window shows the current level of section via a straight line over a reference image. By moving this line, the user can choose a new level of section and therefore a new image. This is done using the control bar to the right of the reference image. Using the arrows the student can move to either the previous, next, first or last section. Using the heavy black pointer, one can move rapidly to any available section. While moving the pointer with the mouse button down, NeuroDatabase shows the location of the available section by the presence of a second, flashing line. If the user releases the mouse button, that section will be chosen.
Click here for Picture


Figure 1. NeuroDatabase's principal windows. A. The Series Browser shows four image icons at a time. A scroll bar is used to move through the series. B. The main Image window. Here several graphic overlays associated with this image are displayed. Just to the right of the image center, the pulvinar nucleus of the thalamus has been selected. C. The Overview window can display all of the available section levels and allows display of any image. D. The Structure Info window displays the name of the selected structure.

NeuroDatabase currently has the capacity to display three series at once. That is, there are three Series Browsers. However, NeuroDatabase can have information about a large number of image series, limited only by disk space. If a series which is not in any of the three browsers is desired, it must replace a currently displayed series. The user can choose the desired series from a list and also choose the destination browser for the series to be displayed.

There are four image series currently in NeuroDatabase. All are from the teaching collection of Harvard Medical School. The first three were originally collected by Paul Yakovlev at Harvard. These images total approximately 35 megabytes. Since only a small number of images are memory resident at one time, NeuroDatabase could, in theory, provide access to hundreds of megabytes of image material.

YC5 - Coronal cross-sections of the human forebrain and brainstem. Myelin stain. 40 images.

BHY6 - Horizontal cross-sections of the human forebrain and brainstem. Myelin stain. 15 images.

YS - Sagittal cross-sections of the human forebrain and brainstem. Myelin stain. 34 images.

Surface - External views of the gross brain surface. 4 images.

Graphic Overlays

A second major feature of NeuroDatabase is the ability to draw in color on top of images, thereby creating graphic overlays. Each image can have any number of graphic overlays which indicate the position or borders of neuronal structures. Graphic overlays are stored separately from the image and can be displayed or hidden, singly or in combination, as needed. There are three kinds of graphic


overlays (splines, irregular polygons and symbols) and two kinds of labels.

For tracing borders of neuronal structures, NeuroDatabase directly supports Bezier spline curves. Compared to irregular polygons, splines make it easier to produce a desired shape, are easier to edit and scale more smoothly. They also provide the best tradeoff between data economy and detail. Symbols, chosen from a large symbol set, can be placed at any desired point to indicate structures where spline contours are not ideal. Symbols (either singly or in groups) are useful for indicating very small structures, structures with ill-defined borders or structures which are enclosed by or overlap other structures.

All graphics can optionally have a text label containing the name of the structure. The instructor can control the size, color, font and placement of this label. NeuroDatabase automatically draws an indicator line from the nearest corner of the label to the graphic overlay. Since symbols are invariably small and thus hard to click on with the mouse cursor, interaction with symbol graphics is accomplished via a specialized label called a "handle". This handle is a square box containing no text, which is connected to the symbol via an indicator line. A mouse click on the handle indicating a structure is equivalent to a mouse click on the structure itself.

Getting Information About Brain Structures

In keeping with the image-centered nature of NeuroDatabase, a student can find out more about a brain structure by clicking on it with the mouse cursor. More specifically, the overlay graphics are the entry to structural information. Typically, when the graphics are visible, clicking on a structure opens the Structure Info window which shows the name of the selected structure. In addition, the graphic overlay is filled in with a stipple pattern to indicate the selection. This is shown in Figure 1B and 1D.

To the right of the structure name in the Structure Info window are two buttons (+ and -) which reveal either more or less information. When only the name is shown, clicking on the + button opens the window to show connection information. The afferents and efferents (inputs and outputs) of the structure are available by clicking on the appropriate button. Clicking on the + again, opens a subsidiary window (Structure Notes) which contains a list of brief clinical/functional correlations for the structure. Figure 2 shows the three states of the information windows.
Click here for Picture

Figure 2. The Structure Info and Structure Notes windows. A. When a brain structure is selected, only its name is shown. B. Clicking on the + button opens the window so the afferents and efferents are available. Here the efferents of the substantia nigra are shown. C. Clicking on the + again reveals the clinical/functional correlations for this structure.

Self-Tests

Four self-tests allow students to assess their own knowledge and practice for exams.

Structure Names - presents a list of the structures in the current section, indicates one structure at a time and asks the student to click on its name in the list.

Structure Borders - (essentially the opposite of Structure Names) presents a list of the structures in the current section, highlights one name at a time and asks the student to click on the structure in the image.

Connections - Tests knowledge of either afferents or efferents (but not both at the same time) of a randomly selected structure in the image. Presents five structures that may be afferents (or efferents) of the selected structure. The student must check the box next to each correct afferent (or efferent).

Correlations - Presents five possible clinical/ functional correlations for the selected structure. The student must check the box next to each correct clinical/functional correlation.
Click here for Picture

Figure 3. The Structure Names self-test. The student must click on the name of the highlighted structure (corpus callosum). Clicking on the Next button randomly selects another structure in this image. The Structure Borders test looks similar.
Click here for Picture


Figure 4. The Connections self-test. In the center right of the image, the anterior nucleus of the thalamus has been selected. The student must check the boxes next to all structures which are afferents of this one (2, 3 and 5). The mechanism will also generate a test for efferents. The Correlations self-test looks similar.

The essential feature of our self-test mechanism is that it creates randomized questions from the database. This randomization can be relatively simple or complex. In the case of the structure names test, the order in which the structures are presented is randomized for each test. Each structure is presented once and when all have been presented, the student is given the option to do the test again. The order is randomized once more and the process is repeated.

In some cases, several types of randomization are used to create a question or set of questions. The Connections test creation routine, for example, goes through the following steps: 1) it chooses a target structure in random order as above. 2) it chooses randomly whether it will test on afferents or efferents. 3) it creates a random answer key for the test. Since there are five possible answers the key is a string of five binary digits (e.g. 11001 or 00010). 4) based on the number of decoy answers needed (positions where there is a 0) it chooses either 1, 2 or 3 decoy structures. These structures are chosen randomly from the entire set of structures which have afferents or efferents defined. 5) For both target and decoy answers, the particular answers which are placed in a given physical slot are randomized.

Gathering User Data

At the end of each session, if data reporting is enabled, NeuroDatabase writes the following information to an ASCII file. If the file already exists, the data for the current session is placed at the end.

User category
User name
Date
Time
Duration of session
(as many lines as necessary, 1 line per self-test)
Self-test test-type current image time spent in test.

Use in 1993

At U. Mass. Medical School, the Neuroanatomy course is given to first-year medical students in the spring term. In 1993 the first-year class had 100 students. Five Macintosh computers (3 Centris 650, 2 Quadra 700) were freely available in the medical library. All were connected via a local ethernet network to 2 Quadra 800 servers. One, the database server, ran Oracle Server for Macintosh (Oracle). The second, the file server, ran Appleshare 3.0 (Apple) and held all digital images.

Student usage data was taken during a six week period. The program was in use for a total of 289 hours, giving a daily average of 6.9 hours. Use of the program peaked in the week before the final exam. Each of the three days before the exam showed over 30 usage hours (5 machines were available, so this was still much less than round-the-clock usage). Total number of self-tests was 3755. As explained above, structure identification for a given image constitutes one self-test. When the student moves to another image and request a self-test, it is a separate self-test. Thus, if all 93 images were studied equally, each would have been used in a self-test 40 times.

The faculty for the course (two of the present authors - MKW and SBG), noted one major change in student behavior. Unlike their experiences in previous years, when it came time to review for the final exam, the students did not want to take time in review sessions to practice identification of structures. This, they said, "we can do on the computer". Instead they wanted to review clinical diagnosis and reasoning about neurologic localization which are best learned through class discussion. As a result, faculty time was more effectively utilized and student-faculty interactions were more interesting for both parties.

At the end of the semester, the students were asked to fill out a written evaluation which covered all aspects of the course. One question asked for the "Strengths of the course, including any extraordinary learning experiences you had." Of 78 responses, 12 mentioned NeuroDatabase specifically as one of the most helpful aids to learning in this course. One question asked about the program itself. "What did you find to be the strengths/weaknesses of ND? What would you like to see added to the program next year?" Out of 64 responses, 7 students were either indifferent (because they did not use the program) or negative because they had problems using it. However, the vast majority of students were extremely positive and asked for NeuroDatabase to include more images, information about connections and function, additional self-tests and labels more suitable for small structures. All these requests are reflected in capabilities described above, either already added in 1993 - 1994 or currently under development.

Discussion

The philosophy that has guided this development is that we wanted first to do those things which are uniquely enabled by computers, rather than those things which are already done well in other media. In our view those are: 1) Providing access to a very large number of high-quality images, many times more than are available in print. 2) Providing rapid access to several planes of section. 3) Providing schematic graphics directly on the image but only as needed. In printed atlases, either a large number of labels are placed on the image, interfering with image details and overwhelming the beginning student, or the images and schematic graphics are placed on separate facing pages requiring continual cross-comparison. 4) Making these schematic graphics interactive. In NeuroDatabase, graphic overlays respond to direct manipulation either for exploration or self-testing.

Use of an underlying relational database has several advantages. In addition to providing specific information about the nervous system, we can provide both general mechanisms for adding and modifying these data and easy methods for their retrieval. We are able to add a very large amount of data without noticeably affecting retrieval performance. We can use a network server for better use of hardware resources, increased communication between users and faculty, and simplification of system management. The database allows us to keep the bookkeeping burden on faculty to a minimum. For example, since each self-test is constructed automatically from database information when the user requests it, a faculty member can add or modify graphical overlays at any time and all self-tests will still be current without additional work.

Our approach to testing is unusual in that each test is generated automatically at the time the student requests it. There are no questions as such, merely several question-generating routines which draw information from the database. Eventually, the routines will create all possible questions; that is, they will traverse the entire data space. This approach has one disadvantage. For all but the simplest identification tests, it would be difficult in our system to measure the difficulty or effectiveness (performance over time) of a particular question, since the exact situation (a combination of question, number of correct answers, actual correct answers and decoy answers) will very rarely be repeated. The major advantage is that instructors can spend all their time adding to or refining the data in the system (graphic overlays, connections and clinical correlations), rather than making up new questions. Once new data are added, a question which involves those data will eventually be generated automatically.

Although we track student usage, we have no reliable data on individual students, since we don't require individual passwords and we encourage group learning. Users are asked to click on their category (e.g. first-year student, medical resident, faculty) and name when they enter NeuroDatabase, but our data analysis pools all student data. We only use the category information to exclude users who are not first-year students from the analysis. Often two or three students will use the program together, so our rough estimate of student contact hours is twice the program usage hours or 578 hours. While we tell the students that we are keeping usage data (with their permission), we are explicit that we are not scoring the self-tests and that performance on the self-tests has no impact on the student's grade.

There are several other neuroanatomy teaching programs either commercially available or in development. BrainStorm, developed by Coppa and Tancred at Stanford University, has very good image quality and series overview capability considering that it is optimized for 13" color monitors. However, the number of images provided and number of self-test questions is relatively low. HyperBrain (S. Stensaas et al., University of Utah) uses videodisc images displayed on a separate video monitor to supplement black-and-white images on a computer monitor. The advantage of this design is that one side of a videodisc can hold 54,000 still images or 30 minutes of moving video (or any combination of these), all of which are available very rapidly (within 1 or 2 seconds). The disadvantages are that structures in the video image are not available for direct manipulation and the user must often cross-compare the black-and-white labeled image with the unlabeled videodisc image. Also, the highest resolution afforded by the videodisc is low to medium-quality compared with digital images made with commonly available technology. NYU Neuroanatomy (K. Walton and D. Hillman, New York University Medical Center) is optimized for 19" color monitors and has excellent image detail. It also has an innovative method of visualizing a gross brain dissection in two planes simultaneously.

While all of the above programs are theoretically modifiable for local needs, in practice they require the person who wants to add data to have complete familiarity with the programming systems used to create them (HyperCard or SuperCard) in addition to neuroanatomy. In contrast, NeuroDatabase was designed for easy data modification and enhancement by non-programming users. To our knowledge, the only other anatomy program which has this feature is Digital Anatomist Browser (6). There does need to be one person at each NeuroDatabase user site who understands the basics of relational databases (and networking if servers are used) and can install and backup the system. This will usually be someone other than the faculty member. Thus, while the programs mentioned above are simple to set up and maintain, they are impractical to customize or enhance. The time investment that NeuroDatabase requires is directly related to its flexibility.

Future development planned for NeuroDatabase includes addition of two new image series (human brainstem and spinal cord) along with their overlay graphics, and a set of "Guided Tours" of several functional systems (e.g. motor system, visual system) Each Guided Tour will be composed of a number of subtopic pages (e.g. motor cortex, extrapyramidal motor system, spinal cord reflexes) accompanied by a list of illustrations. By following through the text and illustrations in order, the user will be introduced to the topic. This is an application of NeuroDatabase which is similar to an electronic book, however the time investment required to make a Guided Tour will be modest since the illustrations will be selected from those labeled images already available in the system.

References

1. Wertheim, S.L. (1989) The Brain Database: a multimedia database for neuroscience research and teaching. Symposium on Computer Applications in Medical Care, 13th Annual Meeting IEEE Computer Society Press. 399-404.

2. Wertheim, S.L. (1990) Analysis and exchange of multimedia laboratory data using the Brain Database. Symposium on Computer Applications in Medical Care, 14th Annual Meeting IEEE Computer Society Press. 389-393.

3. Wertheim, S.L. and Sidman, R.L. (1991) Databases for Neuroscience. Nature 354:88-89.

4. Wertheim, S.L. and Sidman, R.L. (1991) NeuroDatabase: a multimedia neuroscience database for research and teaching. Soc. Neurosci. Abstr. 17:518.

5. Wertheim, S.L. and Sidman, R.L. (1992) Enhancements to NeuroDatabase for teaching. Soc. Neurosci. Abstr. 18:189.
6. Brinkley, J.F., Eno, K. and Sundsten, J.W. (1993) Knowledge-based client-server approach to structural information retrieval: the Digital Anatomist Browser. Computer Methods and Programs in Biomedicine 40:131-145.
Acknowledgements

Supported in part by an Innovations in Medical Education Grant (UMMS) to M.K.W. and S.B.-G. and NIH Grant NS02820 to R.L.S.