Background: Traditional methods for discrimination iron deficiency anemia (IDA) and β-thalassemia trait (BTT), which using CBC indices, are not accurate enough and complementary tests such as Hb electrophoresis are time consuming and expensive. In this study, we introduced the methods with higher accuracy.
Materials and methods: In this study, 510 CBC samples were collected from several screening centers in north of Iran. The number of samples associated with IDA, BTT, and normal subjects were 167, 132, and 211, respectively. The collected samples were used to establish the methods, adaptive neuro fuzzy inference system (ANFIS) and multi-layer perceptron (MLP), through the use of 10-Fold cross validation. In each step of cross validation mathematical methods such as MI, E&FI, S&BI, S&LI, G&KI, EI and SI were investigated by the test samples.
Results: Several indices, such as sensitivity (SENS), specificity (SPEC), positive predictive value (PPV), negative predictive value (NPV), accuracy (ACC), and Youden’s index (YI), have been obtained for the all mentioned methods in each step of Cross Validation. T test showed that the ANFIS and the MLP had not difference (p<0.05). The mathematical methods had not difference (p<0.05), but there was difference between AI-based and Math-based methods (p<0.05).
Conclusion: This study indicates that using artificial intelligence as medical diagnostic tools can help the physicians in discrimination between similar diseases and also it increases accuracy in difficult cases.
Keywords: Iron deficiency anemia, Thalassemia trait, Adaptive neuro-fuzzy inference system, Multi-layer perceptron, Complete blood count.
Khaki Jamei M, Mirzaei Talarposhti K. Cross validation of artificial intelligence and mathematical relationships in the diagnosis of iron deficiency anemia and thalassemia in screening centers of northern Iran in 2014 . MEDICAL SCIENCES 2017; 27 (1) :46-52 URL: http://tmuj.iautmu.ac.ir/article-1-1218-en.html