100 likes | 231 Views
Gene based diagnostic prediction of cancers by using Artificial Neural Network. Liya Wang ECE/CS/ME539. Introduction.
E N D
Gene based diagnostic prediction of cancers by using Artificial Neural Network Liya Wang ECE/CS/ME539
Introduction • Neuroblastoma (NB) , Burkitt lymphomas (BL), Rhabdmyosarcoma (RMS) and Ewing family of tumors (EWS) are different cancers with similar appearance on routine histology. • Accurate diagnosis is essential because the treatment options vary widely depending on the diagnosis. • Gene-expression technique provides a new support for diagnosis.
Project Description • There are 63 training samples and 20 testing samples. Each sample is expressed by 96 genes. Nature Medicine , 7(6), June 2001 • Generalized Hebbian Algorithm (GHA) is used for extracting the principal components of the gene data. • Multilayer Perceptron with Back-propagation learning is responsible for performing the actual classification.
y1 y2 yn Neural Network Design GHA MLP x1 1 x2 x3 xm m n, m=96,n=10
Generalized Hebbian Algorithm • Initialize the weights of the network, wji, to small random values at time k=1. Assign a small positive value to the learning-rate parameter . • Calculate • Increment k by 1, go to step 2, and continue until wji reach their steady-state values.
Multilayer Perceptron • 10 input neurons, 4 output neurons, no hidden layer. • Output neurons use sigmoidal active function • Output from output nodes for training samples are scaled to: [0.2 0.8] • Use the entire training set to estimate training error • # training samples = epoch size = 63