RELAPSE RISK ESTIMATION IN CHILDREN WITH ACUTE LYMPHOBLASTIC LEUKEMIA BY USING NEURAL NETWORKS / ESTIMAÇÃO DO RISCO DE RECIDIVA EM CRIANÇAS PORTADORAS DE LEUCEMIA LINFOBLÁSTICA AGUDA USANDO REDES NEURAIS

AUTOR(ES)
DATA DE PUBLICAÇÃO

2004

RESUMO

In this it is proposed a methodology, based on quantitative procedure, to estimate the adverse event risk (relapse or death) in Acute Lymphoblastic Leukemia (ALL) in children. This methodology was implemented and analyzed in a dataset composed by children diagnosed and treated at the hematology service of the Instituto de Puericultura e Pediatria Martagão Gesteira (IPPMG) in the Federal University of Rio de Janeiro and of the Hospital Universitário Pedro Ernesto (HUPE) in the University of state of Rio de Janeiro. This group constitutes a considerable fraction of the ALL cases in childhood registered in the last few years in Rio de Janeiro. The relapse risk was estimated by a Neural Networks model after a sequence of variable pre-treatment procedures. This treatment has a fundamental importance due to the small number of cases (an intrinsic characteristic of this problem). Although, the ALL is the most frequent cancer in childhood, it incidence is approximately just 1 case for 100 000 inhabitants by year. The obtained results may be considered excellent when compared with the classical risk estimative method used in the medical clinics (BFM risk). A perceptual of successes of 93% (out-of-sample) in no- relapse patients was achieved. We expect that the obtained results may subsidize medical conduct concerning the risk of adverse event and so it could be useful in the treatment intensity modulation.

ASSUNTO(S)

neural networks mutual information classificacao informacao mutua selection of variables risk leucemia linfoblastica aguda redes neurais risco acute lymphocytic leukemia classification selecao de variaveis

ACESSO AO ARTIGO

Documentos Relacionados