Using an artificial neural network to predict traumatic brain injury

Restricted access


Pediatric traumatic brain injury (TBI) is common, but not all injuries require hospitalization. A computational tool for ruling in patients who will have a clinically relevant TBI (CRTBI) would be valuable, providing an evidence-based way to safely discharge children who are at low risk for a CRTBI. The authors hypothesized that an artificial neural network (ANN) trained on clinical and radiologist-interpreted imaging metrics could provide a tool for identifying patients likely to suffer from a CRTBI.


The authors used the prospectively collected, publicly available, multicenter Pediatric Emergency Care Applied Research Network (PECARN) TBI data set. All patients under the age of 18 years with TBI and admission head CT imaging data were included. The authors constructed an ANN using clinical and radiologist-interpreted imaging metrics in order to predict a CRTBI, as previously defined by PECARN: 1) neurosurgical procedure, 2) intubation > 24 hours as direct result of the head trauma, 3) hospitalization ≥ 48 hours and evidence of TBI on a CT scan, or 4) death due to TBI.


Among 12,902 patients included in this study, 480 were diagnosed with CRTBI. The authors’ ANN had a sensitivity of 99.73% with precision of 98.19%, accuracy of 97.98%, negative predictive value of 91.23%, false-negative rate of 0.0027%, and specificity for CRTBI of 60.47%. The area under the receiver operating characteristic curve was 0.9907.


The authors are the first to utilize artificial intelligence to predict a CRTBI in a clinically meaningful manner, using radiologist-interpreted CT information, in order to identify pediatric patients likely to suffer from a CRTBI. This proof-of-concept study lays the groundwork for future studies incorporating iterations of this algorithm directly into the electronic medical record for real-time, data-driven predictive assistance to physicians.

ABBREVIATIONS ANN = artificial neural network; AUC = area under the curve; CRTBI = clinically relevant TBI; EMR = electronic medical record; NPV = negative predictive value; PECARN = Pediatric Emergency Care Applied Research Network; ROC = receiver operator characteristic; TBI = traumatic brain injury.

Downloadable materials

  • Supplemental Material (ZIP 474 KB)

Article Information

Correspondence Andrew T. Hale: Vanderbilt University School of Medicine, Nashville, TN.

INCLUDE WHEN CITING Published online November 2, 2018; DOI: 10.3171/2018.8.PEDS18370.

A.T.H. and D.P.S. contributed equally to this work.

Disclosures The authors report no conflict of interest concerning the materials or methods used in this study or the findings specified in this paper.

© AANS, except where prohibited by US copyright law.



  • View in gallery

    Schematic of the artificial neural network (ANN) constructed here. Seventeen input variables were compared, converging on more than 100 training nodes (for simplicity, fewer nodes are shown). Each input variable connects, analogous to projections in neurons, to each training node. Arbitrary “weights” are then applied to each variable. Each training node is then used to determine the best “weights” of each variable to predict the outcome of interest (“target layer”). Figure is available in color online only.

  • View in gallery

    ROC curve for ANN predictions of CRTBIs. We randomly partitioned patients into 3 groups in order to provide holdout validation on our large data set: 70% were for training the ANN; 15% were for validating the ANN; and 15% were for subsequent final testing of the ANN. The ANN had not been exposed to any of the test patients until after the model was finished training. The ROC for the testing set of patients (left) and the ROC for the entire data set (right). Figure is available in color online only.




All Time Past Year Past 30 Days
Abstract Views 42 42 42
Full Text Views 21 21 21
PDF Downloads 24 24 24
EPUB Downloads 0 0 0


Google Scholar