Compressive Data Storage for Long-Term EEG: Validation by Visual Analysis
Giridhar P. Kalamangalam, Subeikshanan Venkatesan, Maria-Jose Bruzzone, Yue Wang, Carolina B. Maciel, Sotiris Mitropanopoulos, Jean Cibula, Kajal Patel, Abbas Babajani-Feremi
Clinical Neurophysiology Practice
Volume 10
2025
Pages 331-339
ISSN 2467-981X
https://doi.org/10.1016/j.cnp.2025.07.005.
(https://www.sciencedirect.com/science/article/pii/S2467981X25000393)
Abstract:
Objectives: Long-term EEG monitoring (LTM) in acute neurology generates massive data volumes. We investigated whether data-analytic techniques could reduce LTM data size yet conserve their visual diagnostic features.
Methods: LTM exemplars from 50 patients underwent singular value decomposition (SVD). High-variance SVD components were transformed using discrete cosine transform (DCT), and significant elements run-length encoded. Two regimes were tested: (I) SVD and DCT compression ratio (CR) of 1.7 and 12, and (II) CR of 3.7 and 5.7; each achieved an overall CR of ≈20. Compressed data were reconstructed alongside uncompressed originals, to create a total of 200 recordings that were scored by two blinded reviewers. Scores of original and reconstructed data were statistically analyzed.
Results: Score differences between original recordings were smaller than comparisons involving reconstructions using the first regime but did not differ significantly from reconstructions using the second regime.
Conclusions: Raw LTM EEG has sufficient redundancy to undergo extreme (20-fold) data compression without compromising visual diagnostic information. A balanced mix of SVD and DCT appears to be a suitable data-analytic pipeline for achieving such compression.
Significance: Dimension reduction is a significant goal in managing big biomedical data. Our results suggest a pathway for archival of meaningful representations of entire LTM datasets. The latent space suggests new lines of data-scientific inquiry of the EEG in acute neurological illness.
Identification and classification of pathology and artifacts for human intracranial cognitive research
Sarah Long, Maria Bruzzone, Sotiris Mitropanopoulos, Giridhar Kalamangalam, Aysegul Gunduz
NeuroImage
Volume 270
2023
119961
ISSN 1053-8119
https://doi.org/10.1016/j.neuroimage.2023.119961.
(https://www.sciencedirect.com/science/article/pii/S1053811923001076)
Abstract:
Intracranial electroencephalography (iEEG) presents a unique opportunity to extend human neuroscientific understanding. However, typically iEEG is collected from patients diagnosed with focal drug-resistant epilepsy (DRE) and contains transient bursts of pathological activity. This activity disrupts performances on cognitive tasks and can distort findings from human neurophysiology studies. In addition to manual marking by a trained expert, numerous IED detectors have been developed to identify these pathological events. Even so, the versatility and usefulness of these detectors is limited by training on small datasets, incomplete performance metrics, and lack of generalizability to iEEG. Here, we employed a large annotated public iEEG dataset from two institutions to train a random forest classifier (RFC) to distinguish data segments as either ‘non-cerebral artifact’ (n = 73,902), ‘pathological activity’ (n = 67,797), or ‘physiological activity’ (n = 151,290). We found our model performed with an accuracy of 0.941, specificity of 0.950, sensitivity of 0.908, precision of 0.911, and F1 score of 0.910, averaged across all three event types. We extended the generalizability of our model to continuous bipolar data collected in a task-state at a different institution with a lower sampling rate and found our model performed with an accuracy of 0.789, specificity of 0.806, and sensitivity of 0.742, averaged across all three event types. Additionally, we created a custom graphical user interface to implement our classifier and enhance usability.
Mutual Information in Frequency and Its Application to Measure Cross-Frequency Coupling in Epilepsy
R. Malladi, D. H. Johnson, G. P. Kalamangalam, N. Tandon and B. Aazhang
IEEE Transactions on Signal Processing
vol. 66, no. 11, pp. 3008-3023
1 June, 2018
doi: 10.1109/TSP.2018.2821627.
Abstract
We define a metric, mutual information in frequency (MI-in-frequency), to detect and quantify the statistical dependence between different frequency components in the data, referred to as cross-frequency coupling and apply it to electrophysiological recordings from the brain to infer cross-frequency coupling. The current metrics used to quantify the cross-frequency coupling in neuroscience cannot detect if two frequency components in non-Gaussian brain recordings are statistically independent or not. Our MI-in-frequency metric, based on Shannon’s mutual information between the Cramér’s representation of stochastic processes, overcomes this shortcoming and can detect statistical dependence in frequency between non-Gaussian signals. We then describe two data-driven estimators of MI-in-frequency: One based on kernel density estimation and the other based on the nearest neighbor algorithm and validate their performance on simulated data. We then use MI-in-frequency to estimate mutual information between two data streams that are dependent across time, without making any parametric model assumptions. Finally, we use the MI-in-frequency metric to investigate the cross-frequency coupling in seizure onset zone from electrocorticographic recordings during seizures. The inferred cross-frequency coupling characteristics are essential to optimize the spatial and spectral parameters of electrical stimulation based treatments of epilepsy.