Faramarz Valafar

Predicting the Effectiveness of Hydroxyurea in Individual Sickle Cell Anemia Patients

The study described in this paper was undertaken to develop the ability to predict the response of sickle-cell patients to hydroxyurea (HU) therapy. We analyzed the effect of HU on the values of 23 parameters of 83 patients. A Students t-test was used to confirm (Rodgers GP, Dover GJ, Noguchi CT, Schechter AN, Nienhuis AW. Hematologic responses of patients with sickle cell disease to treatment with hydroxyurea, N Engl J Med 1990;322;1037-44) at the 0. 001 level that treatment with HU increases the proportion of fetal hemoglobin (HbF), and the average corpuscular volume (MCV) of the red blood cells. Correlation analysis failed to establish a statistically significant relationship between any of the 23 parameters and the HbF response. Linear regression analysis also failed to predict a patients response to HU. On the other hand, artificial neural network (ANN) pattern-recognition analysis of the 23 parameters predicts, with 86.6% accuracy, those patients that respond positively to HU and those that do not. Furthermore, we have found that the values of only 10 of the 23 parameters (listed in the body of this paper) are sufficient to train ANNs to predict which patients will respond to HU.

Distributed global optimization (DGO)

A new technique of global optimization and its applications in particular to neural networks are presented. The algorithm is also compared to other global optimization algorithms such as the gradient descent method, Monte Carlo method, genetic algorithm and other commercial packages. This new optimization technique proved itself worthy of further study after observing its accuracy of convergence, speed of convergence and ease of use. Some of the advantages of this new optimization technique are given.

Comparative studies of two neural network architectures for modeling of human speech production

A new neural network architecture called the parallel self-organizing consensual neural network (PSCNN) is introduced. Comparative studies of this architecture and the backpropagation (BP) neural network in modeling of human speech production are discussed. The PSCNN consists of a number of self-organizing modules which operate in parallel, during training as well as testing. While these modules can be selected to be any type of neural network, they are chosen to be simple single layer delta rule networks, in this paper. A simplified language, similar to English, is constructed for the purpose of evaluating the performance of the two networks. The behavior of both networks were studied in early stages of training and compared to that of a normal human child in early stages of speech development. In general, the PSCNN network performed considerably better than the BP network in the experiments. Both networks had some errors, but those of PSCNN resembled more closely the human error patterns. In addition, the PSCNN is easy to implement in real-time, parallel architectures. Analysis of certain types of errors indicate that future networks may need some properties of both PSCNN and BP networks.

Parallel, self organizing, consensus neural networks

A neural network architecture, the parallel self-organizing consensus neural net (PSCNN), is developed to improve performance and speed of such networks. The architecture has all the advantages of previous models such as self-organization and possesses new or superior characteristics such as input parallelism and decision making based on consensus. Due to the parallel properties of this network its parallel implementation on an N-cube machine was also studied. The architecture self organizes its modules to maximize performance. Since the system is completely parallel, both recall and learning procedures are very fast. The performance of the network was compared to backpropagation networks in problems of language perception remote sensing and binary logic (Exclusive-Or). PSCNN showed superior performance in all cases studied. In the research reported in the paper, we demonstrate and test the development of the PSCNNs architecture as well as its training rules. In addition, the performance of this new PSCNN system is compared to the performance of backpropagation models.

Artificial neural networks in chemotype analysis of Cryptococcus neoformans

Peak fitting is one of the popular conventional methods for quantitative analysis of /sup 1/H-NMR (proton-nuclear magnetic resonance) spectra in order to establish the serotype or chemotype composition of an antigen. This method of analysis requires human supervision to interpret and manipulate the collected data. Often, due to human error and other factors the results of this analysis are incorrect and unreliable, not mentioning time consuming. A new artificial neural network is developed to automate the same quantitative analysis which previously required human interaction with better precision. ID proton NMR spectra of nearly 100 different strains of Cryptococcus neoformans were used to train and test the network. The results obtained from this network were very comparable and often better than the results of the conventional peak fitting method. The results of neural network however, were produced quickly, without human supervision and thus free of human error.