Predicting developmental disorder in infants using an artificial neural network
Farin Soleimani1, Robab Teymouri, Akbar Biglarian
1Pediatric Neurorehabilitation Research Center, University of Social Welfare & Rehabilitation Sciences,Tehran, Iran. farinir@yahoo.com
Insights
This study developed an artificial neural network (ANN) using perinatal data to predict infant developmental disorders. The ANN model demonstrated superior predictive accuracy and performance compared to traditional logistic regression, aiding early diagnosis.
Area of Science:
- Pediatric Neurology
- Machine Learning in Healthcare
- Developmental Pediatrics
Background:
- Early identification of developmental disorders is crucial for timely intervention.
- Accurate diagnosis is essential to avoid misclassifying healthy children.
Purpose of the Study:
- To develop an artificial neural network (ANN) model for predicting developmental disorders in infants.
- To compare the predictive performance of the ANN model against a logistic regression (LR) model using perinatal data.
Main Methods:
- Recruited 1,232 mother-child dyads from Karaj, Iran.
- Utilized extensive perinatal data, including infant characteristics and medical history.
- Assessed infant development using the validated Infant Neurological International Battery test.
Main Results:
- The ANN model achieved a true prediction concordance of 83.1%, outperforming the LR model's 79.5%.
- Area under the ROC curve for the ANN was 0.79, compared to 0.68 for the LR model.
- ANN demonstrated higher specificity (93.2% vs. 92.7%) and significantly better sensitivity (39.1% vs. 21.7%).
Conclusions:
- Artificial neural networks offer a significantly improved approach for predicting infant developmental disorders compared to logistic regression.
- Perinatal information is valuable for developing predictive models of early childhood development.
- This AI-driven approach holds promise for enhancing diagnostic accuracy in pediatric developmental assessment.
Abstract:
Early recognition of developmental disorders is an important goal, and equally important is avoiding misdiagnosing a disorder in a healthy child without pathology. The aim of the present study was to develop an artificial neural network using perinatal information to predict developmental disorder at infancy. A total of 1,232 mother-child dyads were recruited from 6,150 in the original data of Karaj, Alborz Province, Iran. Thousands of variables are examined in this data including basic characteristics, medical history, and variables related to infants. The validated Infant Neurological International Battery test was employed to assess the infant's development. The concordance indexes showed that true prediction of developmental disorder in the artificial neural network model, compared to the logistic regression model, was 83.1% vs. 79.5% and the area under ROC curves, calculated from testing data, were 0.79 and 0.68, respectively. In addition, specificity and sensitivity of the ANN model vs. LR model was calculated 93.2% vs. 92.7% and 39.1% vs. 21.7%. An artificial neural network performed significantly better than a logistic regression model.
