We use cookies to ensure that we give you the best experience on our website Learn more

Mar 10, 2019

13th World Congress on Brain Injury

Evaluating headache symptoms in persons with a mild traumatic brain injury: The neuronal correlates of self-reported symptoms as features in machine learning analyses

;

machine learning

mTBI

headache

symptom reporting

Abstract

Abstract

thumbnail

Keywords

machine learning

mTBI

headache

symptom reporting

Abstract

Background: Patient reported symptoms in persons with a mild traumatic brain injury (mTBI) are highly subjective and relied upon for diagnostic purposes. Headache and associated pain are frequently reported symptoms following an mTBI and can lead to decreased quality of life. Current research efforts are focused on developing more objective criteria for evaluating persons with an mTBI. Use of supervised machine learning algorithms has demonstrated the capacity to delineate groups using structural and functional brain imaging data. However, the capacity to develop accurate classifiers requires ecologically valid features for analysis. The aim of this investigation was to evaluate the neural correlates of self-reported headache symptoms and determine their capacity to delineate patients with mTBI relative to healthy controls. Methods: A group of patients diagnosed with an mTBI and post-traumatic headache (n=25) were recruited alongside a group of healthy controls (n=11). All participants with an mTBI completed the Rivermead post-concussion scale. Headache scores and total scores were extracted from the Rivermead scale. Cortical thickness was evaluated using Freesurfer and was extracted for 68 brain regions (Whole-brain). Correlation of grey matter thickness was performed with the qdec program from Freesurfer. Regions showing positive and negative correlation with the headache sub-scale of the Rivermead test were extracted and entered as features (Restricted) into machine learning classifiers (k-nearest neighbor (KNN), support vector machine, and Tree algorithms) using Matlab. The Restricted model was evaluated relative to the Whole brain model for accuracy. Results: Participants with an mTBI had a mean of 29.8 (SD=15.6;range:0-65) on the Rivermead symptom scale and a mean of 2.08 (SD=1.35;range:0-5) on the headache sub-scale. A total of 32 clusters were found to correlate with headache symptoms in the mTBI group (p<0.01). Negative correlations were observed in the right inferior and middle temporal area, precentral, and superior parietal area as well as the left fusiform, middle temporal, parahippocampal and lateral occipital cortex. Positive correlations were observed in the right pericalcarine and left lingual gyrus. A total of 24 non-redundant features were submitted to machine learning analysis. Of the models tested, KNN produced the highest accuracy with the Restricted set, delineating mTBI from healthy controls with an accuracy of 67%. This was comparable to use of the Whole brain set that had a maximum accuracy of 70.3% using the KNN method. All other models produced lower accuracy. Discussion: Brain regions found to significantly correlate with headache symptom reporting were distributed throughout the cortex. Positive correlations were found in regions that have previously been observed in migraine patients suggesting a shared neurological focus. Machine learning classifiers based on the restricted features accurately predicted mTBI status similar to the whole-brain approach which suggests that selected features may be highly representative of this cohort.

Discover over 20,000 new abstracts, posters and presentations from leading academic conferences every month. Stay on top of the latest findings, methodologies and discussions happening in your research field around the world.

Company

Legal

Follow us

© Copyright 2019 Morressier GmbH. All rights reserved.

© Copyright 2019 Morressier GmbH.
All rights reserved.