Thursday, May 31, 2001

bookmark: connotea :: del.icio.us ::digg ::furl ::reddit ::yahoo::

Wired News Fingering Cancer Genes
"Genes have fingerprints just like fingers, which got one cancer researcher thinking.

Since the FBI uses neural networks -- a type of artificial intelligence built to imitate neuron function in the brain -- to sift through masses of computerized fingerprint data to solve crimes, why not do the same for genetic fingerprint data?"

""We trained (the neural networks) to recognize this is one cancer and this is another and this is not a cancer," Kahn said. "Eventually it learned to recognize particular features that were particular for cancer.""

EurekAlert Gene chips accurately diagnose four complex childhood cancers
"In this study, for the first time, a gene expression microarray was used to tell the difference between four unique types of cancer: neuroblastoma, rhabdomyosarcoma, non-Hodgkin lymphoma (Burkitt's lymphoma) and Ewing's sarcoma. As a group, these cancers are referred to as the small, round blue cell tumors of childhood because of the way they look under the microscope.

"This is the first time anyone has taken several different kinds of cancer, and used their gene expression patterns for diagnostic classification," Meltzer says. The data form a complex pattern of signal intensities that represent the fingerprint for each tumor type."

"The power of this method, Khan says, "is not only that we can diagnose these cancers, but in the very near future, we will be able to predict which patients will respond to treatment and which will not, and will therefore need stronger treatment.""

Online Journal of Bioinformatics Unsupervised Recognition of Relevant Gene Expression Patterns for Medical Decision Support
"The automated interpretation of gene expression data may play a crucial role in the classification and treatment of human cancers. In this paper a new computational approach to the discovery and analysis of gene expression patterns is illustrated and applied to the recognition of B-cell malignancies. Using cDNA microarrays data obtained from a previous study, an unsupervised and self-adaptive neural network model known as Growing Cell Structures is able to identify normal and diffuse large B-cell lymphoma (DLBCL) patients. Furthermore, it distinguishes patients with molecularly distinct types of DLBCL without previous knowledge of those subclasses."

Pacific Northwest National Laboratory Neural Networks in Medicine
"This document contains references to neural network applications in medicine. It contains some hypertext links to author information and to a few papers. Most of these links are to other WWW, gopher, and ftp sites around the world, and these links may not always be accessable. If a link is repetitively down, please send me a message. Additional links and references are welcome!"

Family Physicians’ Electronic Network Diagnostic Algorithms: results at last!
"We seem to forget, sometimes, that the first researchers in AI that chosen medicine as a problem domain did so, not because of an interest in medicine, but because of an interest in diagnosis as an example of intelligent behavior. Medical diagnosis was one example (perhaps a poor one given that there are much simpler and easier models in other physical systems). Automated diagnosis has rarely interested the medical community, not because of a fear of removing the human element (we've already done that with our reimbursement system) or of replacing humans with machines but, more simply, because diagnosis (as most people view it), is not really the problem. Most clinicians manage some form of diagnosis and most patients are treated appropriately. What is needed is better information on the utility of information and the means to obtain it which least stresses the system. The best source of this information may, in fact, be pooled knowledge of real patients, not compiled knowledge of some particular problem domain.

Of course there are probably not 10 people in NIH who have read Krakauer's article so I don't expect to see any sorely needed policy shifts in NIH funding in the next few years."

Stanford Medical Informatics Preprint Archive Sequential versus standard neural networks for temporal pattern recognition: An example using the domain of coronary heart disease
"Medical researchers who perform prognostic modeling usually oversimplify the problem by choosing a single point in time to predict outcomes (e.g., death in five years). This approach not only fails to differentiate patterns of disease progression, but also wastes important information that is usually available in time-oriented research data bases. The adequate use of time-oriented data bases can improve the performance of prognostic systems if the interdependencies among prognoses at different intervals of time are explicitly modeled. In such models, predictions for a certain interval of time (e.g., death within one year) are influenced by predictions made for other intervals, and prognostic survival curves that provide consistent estimates for several points in time can be produced. We developed a system of neural network models that makes use of time- oriented data to predict development of coronary heart disease (CHD), using a set of 2594 patients. The output of the neural network system was a prognostic curve representing survival without CHD, and the inputs were the values of demographic, clinical, and laboratory variables. The system of neural networks was trained by backprogation and its results were evaluated in test sets of previously unseen cases. We showed that, by explicitly modeling time in the neural net work architecture, the performance of the prognostic index, measured by the area under the receiver operating characteristic (ROC) curve, was significantly improved (p<0.05)."