Artificial Intelligence in Medicine
| Hi-index | 12.05 |
Gestational diabetes mellitus (GDM) is defined as any carbohydrates intolerance recognized first time during pregnancy. Obesity, maternal age, history of miscarriages and life style could be listed as the risk factors. Clinical studies prove that women with gestational diabetes have significantly different pregnancy characteristics and outcomes. For this reason, the possibility of correct diagnosis (especially for women with high risk factors) should be enabled to prevent serious consequences. The purpose of the study was to develop novel GDM diagnosis system based on artificial neural network (ANN). To depict neural data analysis tools, epidemiological database obtained from National Center for Health Statistics (NCHS), Center for Disease Control and Prevention (CDC) was used. Neural analysis was performed using own-written Nets2003 software [Farmacja Polska 54 (1998)]. The procedure was to use ANN as a flexible and advanced tool for modeling relationships between demographic factors and the risk of GDM. Best obtained results (ca. 70% of true positive diagnoses) were compared with logistic regression (56.3 of true positive diagnoses) a classical way of model in case of binary outcome (1-positive vs. 0-negative). The neural model was prepared and after validation and verification procedures was converted into the Java applet, which was subsequently published in the Internet (http://www.cyf-kr.edu.pl/~mfpolak). Such systems are often called 'calculators' and are getting more and more popular in the Web. The Java applet called 'e-ANN' was designed to handle vast variety of neural architectures. ANN is encoded into the ASCII configuration file, which could be easily updated, thus facilitating website management.