Superiority of neural networks over discriminant functions for thalassemia minor screening of red blood cell microcytosis

Arch Pathol Lab Med. 1995 Apr;119(4):350-4.

Abstract

We compared the utility of screening red blood cell (RBC) microcytosis for thalassemia minor using backpropagation neural networks, linear and quadratic discriminant functions, and previously reported discriminant functions based on RBC indices. Screening classification of cases representing possible thalassemia minor (n = 60) and non-thalassemic microcytosis (n = 60) were studied. Among eight RBC indices evaluated, the RBC count was the best univariate discriminant function. Multivariate stepwise discriminant analysis selected the RBC count, the mean corpuscular volume, and the percentage of hypochromic cells as the most discriminatory subset of RBC indices. Optimized linear and quadratic discriminant functions based on these indices performed better than seven previously reported multivariate discriminant functions. However, optimized neural networks were superior to all other discriminant methods studied, averaging 94.1% discriminant efficiency, 94.2% sensitivity, and 94.2% specificity.

MeSH terms

  • Analysis of Variance
  • Discriminant Analysis*
  • Erythrocyte Count
  • Erythrocytes, Abnormal / pathology*
  • Humans
  • Mass Screening
  • Neural Networks, Computer*
  • beta-Thalassemia / blood
  • beta-Thalassemia / diagnosis*