Predictive model of cranial growth in infants with cranial deformation

Differences in cranial growth trajectories define morphological variation in both normal and pathological conditions. Predicting divergent trajectories is key to understanding how and when morphological variation is produced. The incidence of cranial deformation in human infants, specifically deformational plagiocephaly and brachycephaly, has increased dramatically with the initiation of the Back to Sleep campaign in the 1990s. Treatment often requires several months of helmet therapy. However, practitioners have no objective means for reliably estimating the necessary duration of therapy. The goal of this study was to develop a model to provide individualized predictions of duration of helmet therapy using cranial growth trajectories. We obtained STARscanner® 3D laser scan images of 541 children seen in clinic for abnormal head shape at their first visit and again at completion of helmet therapy with IRB approval. Measurements obtained from images at both visits included head circumference, cephalic ratio, oblique diameter difference, cranial vault asymmetry index, sex, and age. A Multivariate Random Forest Model was used to fit the model with 200 trees and 441 patients in the training set. A five‐fold cross‐validation was performed to assess accuracy of the model's prediction of cranial measurements with respect to the duration of helmet therapy. Data from the remaining 100 patients were used to validate the resulting model. Our Multivariate Random Forest model predicts accurately cranial growth trajectories during helmet therapy in infants with cranial deformation. The five‐fold cross validation error of our model is more than 22% lower than standard linear models, with average mean squared error of 0.2568. These results demonstrate that necessary duration of helmet therapy can be accurately predicted using our smart machine learning method. Thousands of infants per year are treated for cranial deformation. Using machine learning, we have developed a model that predicts individualized, evidence‐based treatment strategies accurately. The model is incorporated into a web‐based interface that can be used by practitioners in real time. Used in combination with practitioners' expertise, families will be informed about the likely length of treatment and degree of improvement that may be expected for their individual child. Importantly, while the model developed here is specific to infant cranial deformation, the approach is easily translatable to other craniofacial anomalies, as well as to normal patterns of cranial growth. This abstract is from the Experimental Biology 2019 Meeting. There is no full text article associated with this abstract published in The FASEB Journal .