شبکه عصبی مصنوعی برای ارزیابی خطر اختلالات حرکتی در نوزادان

 Background: Prediction of developmental disorders in infancy is very important. This study aimed to predict movement disorders of children using Artificial Neural Network (ANN) model. Methods: This was a retrospective study, in which 600 infants with normal and 120 infants with abnormal neurologic examination were evaluated. For analysis, the data divided the study group randomly into two equal parts, training and testing. At first the learning process was made on training set (360 cases). After the learning process, testing phase was done with the testing data set (360 cases). All data analysis was carried out by R 14.1 software. Results: For comparing the accuracy of the models' prediction, the accuracy classification table was used for the testing subset. The concordance indexes showed that the ANN model led to more accurate predictions compared to the LR model (true prediction with or without developmental disorder was 78.6% vs. 73.9%). The under Receiver Operating Characteristic (ROC) curves, calculated from testing data, for ANN and LR model were 0.71 and 0.68, respectively. In addition, specificity and sensitivity of the ANN model vs. LR model was calculated 88.0% vs. 85.0% and 31.7% vs. 18.3%, respectively. Conclusions: The ability of ANN and LR predictions to identify infants without developmental disorder is similar but the ability of the ANN predictions to identify infants with developmental disorder is better than LR predictions.