Home » Technology » Scientific Computing » (Artificial) Mental Imagery

(Artificial) Mental Imagery

© Adam Hughes

One of the unifying themes in all areas of scientific computing is the need for increasingly sophisticated imaging techniques. From chemists to biologists to acoustical physicists, we all generate large amounts of data, and we all need workable ways of interpreting them. Probably the most common and desirable method of dealing with this data, since it allows for both qualitative and quantitative analysis, imaging is absolutely essential in some fields of research. A former colleague of mine who employed electromagnetic and acoustical methods in military vehicle identification used to drive home this point : "We can have all the data in the world about an enemy aircracft, but if we don't have an accurate image to show us where the wings, cockpit, etc., are, the it doesn't really make much difference."

In the search for better and better imaging methods, scientists have begun to explore the possible use of neural networks in this area. Because of the importance of imaging in the medical field and the resultant wealth of images from which to run tests on new methods, researchers at Battelle Memorial Institute in Richland, WA, have undertaken a massive effort in applying neural networks to medical imaging.

The Battelle team is employing a neural network that is slightly different in setup than the ones used for the artificial noses we examined last week. The Pulse-Coupled Neural Network (PCNN) is modeled after the mammalian visual cortex and is based on technology developed to explain the pulse synchrony known to exist in the feline visual cortex. In this model, a neuron is connected to an image pixel, or a set of image pixels. The neuron processes signals from both the pixels and its neighboring neurons. Similar pixels cause the associated neurons to fire simultaneously, and the synchronization of the resultant pulses can be used to group similar structures in the image.

Because of the features described above, it has been found that PCNN's are potentially useful toools for edge detection and feature separation in certain types of images. Next week, we'll look in a little more detail at the specific problems examined by the Battelle research team in their efforts to apply neural network technology to medical imaging.

The copyright of the article (Artificial) Mental Imagery in Scientific Computing is owned by . Permission to republish (Artificial) Mental Imagery in print or online must be granted by the author in writing.