534 - Predicting Neurodevelopmental Outcome in Very Low Birth Weight Infants from MRI Utilizing a Machine Learning Model with Volumetrics Extracted from Infant Freesurfer.

Poster Number: 534
Publication Number: 534.1481

Background: Infant Freesurfer was introduced several years ago to address unmet needs in the highly specialized infant brain. This automated algorithm made volumetrics, previously research tools requiring specialist infrastructure and expertise, more accessible to clinicians.

Objective: The purpose of this study is to develop a new model for predicting the long-term neurodevelopmental outcomes of preterm infants using automated volumetry extracted from term-equivalent age MRI, diffusion tensor imaging, and clinical information.

Design/Methods: Preterm neonates hospitalized at Severance Children’s Hospital, born between January 2012 and December 2019, were consecutively collected. Inclusion criteria included infants with birth weights under 1500 g who underwent both term-equivalent age (TEA) MRI and Bayley Scales of Infant and Toddler Development 2nd Edition (BSID-II) assessments at 18 months of corrected age (CA). The study utilized the random forest classifier and logistic regression methods to develop three models. After development, the superior model was applied to the test set, and various performance metrics, including the area under the receiver operating curve (AUROC), accuracy, sensitivity, precision, and F1 score, were evaluated.

Results: A total of 150 patient data were enrolled based on the defined criteria. For predicting low Psychomotor Development Index (PDI), the random forest classifier was employed. The AUROC values for models using clinical variables, MR volumetry, and both clinical variables and MR volumetry were 0.8435, 0.7281, and 0.9297, respectively. To predict low Mental Development Index (MDI), a logistic regression model was chosen. The AUROC values for models using clinical variables, MR volumetry, and both clinical variables and MR volumetry were 0.7483, 0.7052, and 0.7755, respectively. The model incorporating both clinical variables and MR volumetry exhibited the highest AUROC values for both PDI and MDI prediction.

Conclusion(s): In conclusion, a new prediction model utilizing an automated volumetry algorithm significantly distinguishes the long-term psychomotor developmental outcomes of preterm infants. This model can provide additional information in clinical settings without requiring specialized expertise.