Machine Learning Algorithms and Quantitative Electroencephalography Predictors for Outcome Prediction in Traumatic Brain Injury: A Systematic Review

Abstract

Recent developments in the field of machine learning (ML) have led to a renewed interest in the use of electroencephalography (EEG) to predict the outcome after traumatic brain injury (TBI). This systematic review aims to determine how previous studies have taken into consideration the important modeling issues for quantitative EEG (qEEG) predictors in developing prognostic models. A systematic search in the PubMed and Google Scholar databases was performed to identify all predictive models for the extended Glasgow outcome scale (GOSE) and Glasgow outcome scale (GOS) based on EEG data. Fourteen studies were identified that evaluated ML algorithms using qEEG predictors to predict outcome in patients with moderate to severe TBI. In each model, a maximum of five qEEG predictors were selected to determine the association between these parameters, and favorable or unfavorable predicted outcomes. The most common ML technique used was logistic regression, but the algorithms varied depending on the types and numbers of qEEG predictors selected in each model. The qEEG variability for the relative and absolute band powers were the most common qEEG predictors included in the models (46%) followed by total EEG power of all frequency bands (31%), EEG-reactivity (31%) and coherence (15%). Model performance was often quantified by the area under the receiving-operating characteristic curve (AUROC) rather than by accuracy rate. Various ML models have demonstrated great potential, especially using qEEG predictors, to predict outcome in patients with moderate to severe TBI. INDEX TERMS Electroencephalography, machine learning algorithms, prediction methods, reviews.

Improving Outcome Prediction for Traumatic Brain Injury From Imbalanced Datasets Using RUSBoosted Trees on Electroencephalography Spectral Power

“…Several studies have proposed that there could be a new strategy to develop a robust predictive model for predicting (i.e., favorable or unfavorable) outcomes after TBI based on qEEG features [16]- [20]. Despite the difficulties, recent years have seen the successful development of a predictive model, using ML algorithms and different qEEG features, which then has decreased human error and provided a better TBI prognosis [21], [22].…”

Regression Networks for Neurophysiological Indicator Evaluation in Practicing Motor Imagery Tasks

“…In this regard, the correlation between the neural activity features that are extracted in advance (electrophysiological indicators or predictor) with the MI onset responses instructed via sensory stimuli can be assessed to prescreen participants for the ability to learn regulation of brain activity (pre-training measures) or for the improvement of learning abilities (training phase) [ 17 ]. A systematic review of the predictors of neurofeedback training outcome is given in [ 18 , 19 ], concluding that the most promising predictor seems to be the (neurophysiological) baseline activity, which was derived from the parameter targeted by the training. In an attempt to anticipate the evoked MI responses, several pre-training electrophysiological indicators are reported, like functional connectivity of resting-state networks [ 20 ], rhythm activity of eyes-open and eyes-closed resting-states [ 21 ], pre-cue EEG rhythms over different brain regions [ 22 ], and the power spectral density estimates of resting wakefulness (before the cue-onset of the conventional MI trial timing and resting state) [ 23 , 24 ].…”

A New Approach to High-Order Electroencephalogram Phase Analysis Details the Mathematical Mechanisms of Central Nervous System Impulse Encoding

“…The index of coherence, that measures the degree of phase similarity and thus functional connectivity for different brain locations, as well as long-and short-range correlates between frequency band components for different brain locations, also make valuable contributions to the understanding of brain function (e.g., by demonstrating that smaller amplitude alpha activities in prefrontal and frontal locations are not solely due to the conduction of electrical currents associated with typically larger amplitude alpha activities in occipital-parietal locations, as was once supported) [11,24,25]. Relatively simple indices that are not specific to any particular band include median frequency and spectral edge frequency (SEF), where SEF is defined as the frequency below which a specified percentage (typically 95%) of a spectrum's power exists [23,26].…”

“…Frequencydomain spectra provide a range of complementary diagnostic indices which often refer to the traditional delta, theta, alpha and beta bands, approximately corresponding to spectral frequency ranges of 0.5 to 4, 4 to 8, 8 to 13 and >13 Hz respectively (an additional gamma band that caps and surpasses the beta band from approximately 30 Hz onwards may also be included for relatively high frequency spectra). The total and proportional activities (or powers) of the traditional bands, as well as several permutations of ratios of band activities are representative of conventional EEG indices that are monitored within a diagnostic setting [11,[21][22][23], with related but more advanced indices formed by combining several frequency band components (e.g., of the SSVEP).…”

“…Functional magnetic resonance imaging (fMRI) and positron emission tomography (PET) increasingly offer competing alternatives or complementary options to EEG analysis, as they collectively detect parameters such as: (i) cerebral blood flow and fluctuations in capillary deoxyhaemoglobin populations which are triggered by the activity of neural populations; (ii) the rate of glucose metabolism which is proportional to the rate of synaptic firing; and (iii) enhanced dopamine secretion which is linked to the activation of genes responsible for brain function [6][7][8][9][10][11]. For example, through the detection of otherwise finely hidden levels of brain activity, fMRI and (especially) PET offer improved differential diagnosis and long-term recovery likelihood prognostication within consciousness disorders such as unresponsive wakefulness syndrome [8].…”