Next: Network Up: Title Page Previous: Segmenting Index: Full Text Index Contents: Conference Page

Front-End to Large Digital Libraries

One of the most intriguing possibilities under consideration, is that of the Visible Human Explorer as a visible, interactive window into a database of textural, numerical, and additional visible data. Fast access to an imagery database leads us to begin to think of them in ways analogous to text-based databases and their potentials and limitations. For example the dichotomy between a broad search to be sure to get the material you want and a very narrow search to avoid getting too much. Various computer algorithms are being developed to allow intelligent searches of imagery databases and the Visible Human Explorer is compatible with that. However, an additional advantage is present. Evolution has left us with an excellent ability to recognize patterns, and in turn the absence of a pattern, that anomalous item in a sequence. Normal human activities develop incredibly efficient hand-eye coordination and also incredibly quick pattern recognition. These qualities, linked with this browsing software, form the basis of a powerful front end to various data sets. The software provides both fast random and fast sequential access to large data sets. A researcher can scan through the data with images flashing on the screen in front of his eyes. When the eye/brain recognizes a single image of interest out of the flashing sequence, the brain/hand can stop the sequence. If the sequence is not stopped in time it can be run backwards from that point, under control of mouse movement to the image of interest. Information linked with that image could then be displayed. Here that might be the CT or MRI scan associated with the same section as the photographic image. An extension of that is that the cursor position on that particular image may provide a link to a great deal of information about that voxel location - the muscle type, CT density, etc.

The linked text database of 1500 anatomical structure names integrated in the Visible Human Explorer is a prototype of a wider application of such database linkages. With such linked data sets, the Visible Human Project images may become either the initiators or the result of a user's search. Image access is direct and interactive. There is no need to previously plan and choose images to load and wait during a lengthy image loading stage before the pixel location can be found. In a simple case, upon entering "spinal cord", the system automatically prepares a list of sequential image filenames containing the spinal cord. This list in turn drives the animation of only those images. Note that animated images do not need to be stored sequentially on the disk to maintain display speed from hard disk. By linking the database to the images through a lookup table, the database can be easily expanded, contracted or replaced completely. All that needs to be changed is the table and storage disks as appropriate.

In the other direction, the user may click on an individual voxel, individual image, or range of images and be returned information pertinent to the input, the associated muscle name, the mechanical resistance, and the CT density and so on or to a related series of images. In an educational setting, clicking of the CT, X-ray, or colour photograph image of the healthy body of the Visible Human Project can link to a large set of X-rays of real patients with various conditions.

Beyond the Visible Human Project data sets, a doctor may wish to scan a large number of X-rays or other images looking for something similar to the one he has in hand from a patient. He is looking for something similar to the one he has and in turn information on the treatment of that case. He cannot search the other way because the X-ray database has not been indexed to that level, or it is partially indexed and the keyword search returns 300 images. At this point, a customized version of the Visible Human Explorer can take over the list of those 300 images and let him quickly browse them to find the matching image(s).

The 45 gigabytes of raw Visible Human Project data has now become simply a minor storage problem, a few hard disks. Local, immediate access to it is not a challenge. It is an example of the imagery that is coming continuously every day. The University of Michigan Medical School, for example, produces greater than 2 terabytes of microscopic 2-D and 3-D research microscopic digital imagery per year. This led to the consideration of a distribution network linked to a central storage facility.

Next: Network Up: Title Page Previous: Segmenting Index: Full Text Index Contents: Conference Page