Expert-level detection of low-voltage intervals in EEG as a robust biomarker of neonatal encephalopathy

The gist The automated quantification of LVIs provides a robust clinical biomarker of neonatal encephalopathy with clear associations with background EEG activity, clinical HIE severity, and seizure risk.

Introduction and Motivation

Neonatal encephalopathy is a critical marker of brain dysfunction but can be difficult to quantify through visual review It is a leading cause of neonatal death and lifelong neurological impairment The electroencephalogram (EEG) is the clinical gold standard for monitoring brain function and can be used to assess and track evolving injuries and neurological recovery Expertise in interpretation of neonatal EEG is scarce and rarely available 24/7 Artificial intelligence (AI) has the potential to assist with EEG review as it excels at pattern matching tasks such as identifying specific waveforms within the signal The application of AI to neonatal EEG has primarily centred on seizure detection Stratifying the different grades of severity of background activity has the potential to provide a useful metric of overall brain health that is more consistent than the episodic nature of seizures Inter-burst intervals (IBIs), which are periods of low-voltage or suppressed activity between bursts, are the defining characteristic of discontinuity Automated detection of IBIs is challenging because the clinical context dictates interpretation Thus far, existing IBI detection methods have largely ignored this context, training and testing on data from a single vigilance state or background pattern only In this study, we focus on low-voltage activity itself, without requiring a burst–interburst pattern We use the term low-voltage interval (LVI) to describe a period of EEG activity meeting prespecified low-amplitude and duration criteria in an individual channel An LVI does not require identifiable preceding or subsequent bursts and may be confined to a single channel or occur concurrently across multiple channels We develop machine-learning (ML) methods that are trained and validated on a large and diverse dataset encompassing different types of EEG vigilance states and background patterns, comparing with existing state-of-the-art methods We then establish automated LVI detection as a clinically relevant biomarker of neonatal encephalopathy across four applications: 1) grading EEG background on an open benchmark, 2) classifying clinical HIE severity against a healthy baseline cohort, 3) predicting seizures, and 4) correlating LVI features with total seizure burden

Methodology for LVI Detection

The definition of LVIs is established as intervals of activity below 25 µV for 2 or more seconds on an individual channel These events are not tied to specific vigilance states or limited to specific background patterns of activity Unlike IBIs, they do not necessarily require a preceding or subsequent bursts The decision to annotate an event lies with the EEG reader, who uses their experience and judgement to make this determination To develop an automated LVI detection method, we started by exploring existing methods for detecting IBIs We test this hypothesis by comparing feature-based approaches with an existing ML method for detecting IBIs [22] We also develop a neural network model to detect LVIs, specifically using a ConvNeXt architecture [29] The Nano version of the model, a small model with 38.7k trainable parameters, was used as we found no further gains by scaling model size Data augmentation during training included signal inversion (flip), cutout [30], and a time-warping method [2610.12680X>

Detection Performance and Expert Equivalence

For training and validation of the detection methods, 654 h of per-channel EEG from 123 neonates were reviewed, resulting in 51,162 LVI annotations On held-out validation, the ConvNeXt method had an area under the receiver operating characteristic curve (AUC) of 0.996 and a Matthews correlation coefficient of 0.926 The ConvNeXt model agreed more with the individual EEG readers than the EEG readers agreed with each other (∆κ = 0.025, 95% CI: 0.007 to 0.045, p < 0.01) thus attaining expert-level equivalence. Both SVM and ConvNeXt methods reach expert-level agreement (p < 0.01).

Clinical Relevance and Biomarker Potential

We tested the association of LVIs to different markers of neonatal encephalopathy across four applications For the first LVI application, the six features were computed for the 169 1 h epochs (n = 53 neonates) from the open-access dataset with the four grades of background abnormalities The median-duration LVI feature achieved the strongest separation across grades, with ρ = 0.821 and ε2 = 0.676 For the second LVI application, we examined separation among the clinical HIE grades The LVI percentage feature produced the strongest separation across the four grades, with ρ = 0.384 and ε2 = 0.297 In the third application experiment, we examined neonates with seizures LVI percentage as the leading discriminating feature with an AUC of 0.872 and ρ = 0.475 In the fourth application experiment, we evaluated the association of LVI activity with seizure burden For example, maximum duration had a significant correlation with seizure burden (ρ = 0.27) and there was a difference among the four categories of burden (ε2 = 0.15)

Conclusion

Automated quantification of LVIs provides a robust clinical biomarker of neonatal encephalopathy with clear associations with background EEG activity, clinical HIE severity, and seizure risk Objective and repeatable quantitative LVI features are able to stratify EEG background grades as well as clinical grades of HIE, and accurately predict which neonates will later have seizures This demonstrates the clinical relevance and importance of LVIs in characterising pathological EEG activity related to neonatal brain injury

Appendix Details

The amplitude-based features include the peak-to-peak measure, defined in (A.1), which calculates the ratio of activity outside the ±12.5 µV range over a 1 s window The envelope of the signal, defined as E(n) = x(n) + jH[x(n)] (A.2) Other features include the envelope-derivative operator (EDO) [24] and line-length, defined as L(n) = X fs m=1 x(n − m + 1) − x(n − m)

References

[3] S. Tan, S. Mageziyi, L. Long, N. Dilimulati, Z. Hui, and N. Rejiafu “Analysis of global, regional, and national burdens of neonatal encephalopathy from 1990 to 2021: insights from the global burden of disease study 2021” Frontiers in Public Health, vol. 13, Oct. 2025[3]

[4] S. M.

1. Bold header: Expert-Level Equivalence in Detection

The ConvNeXt model was selected for all application experiments because it comfortably achieving expert-level equivalence with SVM, as both methods demonstrated a positive change in Cohen’s kappa (positive values of ∆κ). This indicates that the AI system achieves performance indistinguishable from a human expert for LVI detection.

2. Bold header: Robust Feature Engineering

The study suggests that simple feature-based methods fail to achieve comparable results and implies that a large set of annotations is required to attain expert-level detection. Future systems should incorporate a small, interpretable feature set, such as the six LVI features (maximum and median duration, percentage, number of events), which showed remarkably good discriminatory power across the EEG grades.

3. Bold header: Application in Clinical Stratification

The improved AI system can stratify EEG background grades as well as clinical grades of HIE using objective LVI features. Specifically, the median-duration LVI feature achieved the strongest separation across grades, with ρ = 0.821, allowing for precise classification based on amplitude and duration characteristics.

4. Bold header: Predictive Seizure Modeling

The system can accurately predict which neonates will later have seizures using a high-performing model where LVI percentage [was] the leading discriminating feature with an AUC of 0.872. This allows for real-time risk stratification based on detected low-voltage activity patterns.

5. Bold header: Temporal Trajectory Monitoring

The system can track development or recovery of EEG activity in real time by monitoring LVI percentage, which showed stable separation across the 24 hours for all three grades, enabling continuous assessment of neurocritical status over extended periods.

Background: Discontinuity in neonatal EEG is a critical marker of brain dysfunction but can be difficult to quantify visually. Inter-burst intervals (IBIs) are a key characteristic of discontinuity, yet their definition does not easily translate to AI detection. Objectives: To introduce low-voltage intervals (LVIs), defined exclusively by amplitude and duration, as a context-independent alternative to IBIs, develop an AI detection method for LVIs, and assess their clinical utility as a marker of brain injury. Methods: We developed a ConvNeXt model to detect LVIs and compared its performance with existing methods. We then tested the clinical relevance of detected LVIs across four applications: 1) grading background EEG abnormalities on an open benchmark dataset, 2) classifying clinical hypoxic-ischaemic encephalopathy (HIE) grades and comparing with a normal EEG dataset, 3) seizure prediction, and 4) association with seizure burden. Results: For training and validation, 654 hours of per-channel EEG from 123 neonates were reviewed and 51,162 LVIs were annotated. On held-out validation, ConvNeXt had an AUC of 0.996 and a Matthews correlation coefficient of 0.926. Model-reader agreement exceeded inter-reader agreement (Δκ= 0.025, 95% CI: 0.007 to 0.045, p<0.001), attaining expert-level equivalence. On a combined dataset of 444 neonates, percentage LVI showed a strong association with EEG grades (Spearman's ρ=0.801), a moderate association with HIE grades (ρ=0.384), a weaker association with seizure burden (ρ=0.244), and a strong seizure prediction capacity (AUC =0.872). Conclusion: AI detection of LVIs achieves expert-level equivalence, whereas quantitative features do not. Automated LVI quantification provides a robust clinical biomarker of neonatal encephalopathy associated with background EEG activity, HIE severity, and seizure risk.