A new method to predict the quality of umbilical cord blood units based on maternal and neonatal factors and collection techniques

The saving banks of “umbilical cord blood stem cells” are considered as strategic health-based institutions in most countries. Due to the limited capacity of cord blood sample storage tanks, the samples should be evaluated according to their quality. So these banks need a method to assess quality. In this paper, first, the effective factors on the quality index of the extracted cord blood from newborn infants are identified using the electronic records and database of Royan’s umbilical cord blood bank. Then by machine learning and various statistical methods such as multilayer perceptron neural networks, radial basis function neural networks, logistic regression, and C4.5 decision tree, the quality value of blood samples and their proper category (for discarding or freezing) are determined. Two different sets of data have been used to evaluate the proposed methods. The results show that the ensemble of radial basis function neural network with k-means clustering model has the best accuracy compared to other methods, which categorizes the samples with 91.5% accuracy for the first data set and 81.6% accuracy for the second one. The results also show that using this method can save about $1 million annually.