Artificial Neural Networks in Clinical Biochemistry

Introduction and Goals: Artificial Neural Network (ANN) is a powerful and preferred tool for data mining because of their flexibility and accuracy. The main purpose of this article is to describe the importance of ANN analysis in predicting the relation of biochemical and hematologic indices to erythrocyte count. Methods: Blood samples were collected from women who referred to clinical laboratory. Iron, TIBC, FBS, Urea, uric acid, LDL, HDL, triglyceride, cholesterol, bilirubin, ALT, AST and alkaline phosphatase were assayed as biochemical parameters. Hematologic parameters were HCT, RBC, MCV, Hb, MCH, MCHC, WBC, PLT and ESR. After data extraction, the results were analyzed by artificial neural networks. The characteristics of our multilayer perceptron (MLP) were: a) biochemical and hematologic indices as independent variables in input layer, b) batch method in the training phase, c) standardized method for rescaling the covariates, d) hyperbolic tangent and identity as activation functions, e) scaled conjugate gradient as optimization method, and f) erythrocyte count as dependent variable in output layer. Results: According to our results, independent variable importance analysis showed that Hb concentration (100%), MCH (96.5%), ALT (70.8%), cholesterol (69.4%) and FBS (63.8%) were the most important parameters with the normalized importance values more than 50%. Also sum of squares errors (0.182 and 0.072), relative errors (0.026 and 1.986), and correlation between predicted and observed values (y=0.58+0.88x) showed that our ANN models have acceptable validity in predicting the most important parameters in relation to erythrocyte count.Conclusion: ANN analysis is a powerful, reliable, accurate and important method in different fields of clinical laboratory, especially in complicated processes such as correlation between biochemical and hematological indices.

Artificial Neural Networks in Clinical Biochemistry Keywords:

Artificial Neural Network , Biochemical indices , Clinical Laboratory , Erythrocyte count , Hematologic indices