Sigma Xi Presentation: Implementation of Time Frequency Analysis for Seizure Localization
- 1. Implementation of Time Frequency Analysis for Seizure Localization: Phase II Asha Reddy Lake Highland Preparatory School, Orlando, Florida, USA
- 2. Research Question and Goal Can EEG analysis of absence seizures and modelling of overtone harmonics be improved by replacing spectrogram features with Cohen-Posch features? The goal of this phase of the project is to investigate whether Cohen-Posch time-frequency analysis can provide improved modelling of EEG scans, especially relating to the strength of overtones which are necessary for the accurate localization of absence seizures.
- 4. Absence Epilepsy - Absence seizures are caused by brief abnormal electrical activity in a person’s brain - When people have absence seizures, they are not aware of what is happening around them - Absence seizures are mostly seen in young children - An observer may not see the beginning or end of a seizure - Because of the unpredictability and irregularity of absence seizure, prediction and detection methods are vital
- 5. Electroencephalograms - Epilepsy is generally modelled using electroencephalograms, or EEG scans, which are tests that detect electrical activity in a person’s brain and consist of measurements of a set of potential differences between pairs of scalp electrodes. (Baillet, 2011) - Signals recorded from living neurological tissue are extremely noisy at all scales from individual ion channels through collections of one or more neurons up to scalp-recorded EEG scans. The noise recorded in these EEG’s causes uncertainty in the analysis of seizure data. - It is vital to find an alternative method of analyzing and modeling EEG data to improve the technique of epilepsy diagnosis. Figure 1: https://www.epilepsyqueensland.com.au/about- epilepsy-epilepsy-queensland/seizure- types/what-are-the-different-types-of-seizures
- 6. An Introduction to Time Frequency Analysis and the Cohen-Posch Method - Though the spectrogram is the most well known method of TFA for seizure analysis as well as most other applications, there are multiple limitations. - The spectrogram is “window dependent” making it difficult to separate the effects of the window from the inherent information and characteristics of the signal. - The Cohen-Posch method has the properties of being positive as well as having correct marginals. - Positive time-frequency dimensions are used in various applications such as the analysis of multicomponent signals, biomedical engineering, and speech processing. - Due to the mathematical advantages of this time-frequency method compared to older methods, it is logical and necessary to extend the Cohen-Posch methodology to seizure data. Figure 2: A spectrogram plot from MATLAB eeglab Figure 3: A Cohen-Posch positive time frequency plot (not seizure data).
- 7. Overtone Harmonics - Overtone harmonics emphasize the fundamental or dominant tone in EEG frequencies - This dominant tone acts as the primary seizure generator as it forces oscillations in other parts of the brain - It is a good marker of where absence epilepsy starts in the brain - Comparing overtone harmonics between the Spectrogram and Cohen Posch method is important in deciding whether the Cohen Posch method has improved qualities of EEG detection
- 8. Phase I Summary - In the first phase of this project, public EEG data of absence seizures was converted from a bipolar montage to a monopolar montage, which was necessary in order to test various methods of time- frequency analysis on this data. - The main problem with multipolar montages is that they lose important data that is needed for interpreting the signals. For this reason, the first phase of this project was necessary. - This converted data was used in Phase II. Figure 4: A bipolar montage of absence seizure EEG data Figure 5: A monopolar montage of absence seizure EEG data
- 10. Methodology Overview Investigate the mathematics behind signal processing and learn basic Matlab coding Extract absence seizure EEG datasets from the CHB-MIT database and run spectrogram and Cohen-Posch code. Beta test the spectrogram and Cohen- Posch code in Matlab and debug when necessary Compare overtone harmonic entropies of each method to demonstrate enhanced performance
- 12. Results
- 13. Results: Spectrogram vs Cohen Posch Methods on an Absence Seizure The Cohen-Posch method rescales the spectrogram to emphasize the overtone harmonics of the fundamental seizure frequency (f0) The presence of the strong fundamental seizure frequency and overtones is diagnostic of an absence seizure. The emphasized harmonics allow for a more precise detection of the seizure and more accurate estimate of f0. Figure 6: Spectrogram vs Cohen-Posch figures. The pink and red frequencies are overtone harmonics.
- 14. Results: Comparing Overtones Using Entropy The strength of the overtone harmonics at various times throughout the seizure was compared between the Spectrogram and Cohen-Posch methods using a measure of entropy. A higher entropy represents a higher certainty that the fundamental frequency will be detected. This method of data analysis provided the following figures.
- 15. Pre-Ictal Entropies Figure 7: Pre-Ictal (before seizure) entropies. Top graph shows the overtone entropy of the spectrogram. Bottom graph shows the overtone entropy of the Cohen-Posch method. The entropy ratio is 1.0625. This means there is not a significant difference of overtone strength before the seizure taking place.
- 16. Beginning Ictus Entropies Figure 8: Beginning ictus (start of seizure) entropies. Top graph shows the overtone entropy of the spectrogram. Bottom graph shows the overtone entropy of the Cohen- Posch method. The entropy ratio is 3.8087. The Cohen-Posch is now providing almost 4 times the overtone entropy of the spectrogram.
- 17. Ictal Entropies Figure 9: Ictal (during seizure) entropies. Top graph shows the overtone entropy of the spectrogram. Bottom graph shows the overtone entropy of the Cohen-Posch method. The entropy ratio is 5.4797. The Cohen-Posch is now providing about 5 times the overtone entropy of the spectrogram.
- 18. Interpretation: Comparing Overtones Using Entropy Figure 10: This is a comparison of the overtone entropies between a spectrogram and the Cohen- Posch method. It is clear that between time 2455 and 2470 seconds, the ratio of overtone strength between the two methods clearly increases proving that the Cohen-Posch method does provide enhanced modelling of overtone harmonics.
- 19. Conclusion
- 20. Conclusions and Possible Applications - The results show that the overtone strength of the Cohen-Posch method is enhanced in comparison to that of the Spectrogram. - The emphasized harmonics allow for a more precise detection of the seizure and more accurate detection of this fundamental frequency. - These results may also be used for improving predictive models of absence seizures. If overtone strength is more clearly modelled using the Cohen-Posch methodology, it could be an essential discovery to absence seizure research. This method may be helpful in developing an objective program which can detect active or upcoming absence seizures. - The next phase of this project will be comparing other features of the spectrogram and Cohen- Posch methods to determine whether the Cohen-Posch method can serve as an alternative method of analyzing and modeling EEG data to improve the technique of epilepsy diagnosis.
- 21. References
- 22. Baillet, S. (2011). Electromagnetic Brain Mapping Using MEG and EEG. Oxford Handbooks Online. doi: 10.1093/oxfordhb/9780195342161.013.0007 Davy, M., & Doucet, A. (2003). Copulas: a new insight into positive time-frequency distributions. IEEE Signal Processing Letters, 10(7), 215–218. doi: 10.1109/lsp.2003.811636 Loughlin, P., & Bernard, G. (1997). COHEN–POSCH (POSITIVE) TIME–FREQUENCY DISTRIBUTIONS AND THEIR APPLICATION TO MACHINE VIBRATION ANALYSIS☆. Mechanical Systems and Signal Processing, 11, 561-576. Loughlin, P., & Cohen, L. (2018). Positive Time–Frequency Distributions. Applications in Time-Frequency Signal Processing, 121–162. doi: 10.1201/9781315220017-3 Sanei, S., & Chambers, J. (2013). EEG Source Localization. EEG Signal Processing, 197–218. doi: 10.1002/9780470511923.ch5 So, E. L. (2000). Integration of EEG, MRI, and SPECT in Localizing the Seizure Focus for Epilepsy Surgery. Epilepsia, 41(s3). doi: 10.1111/j.1528-1157.2000.tb01534.x Oppenheim, Willsky, & Navab. Signals and Systems, 2nd Edition, Prentice-Hall Signal Processing Series, Prentice- Hall,1997 Https://www.epilepsyqueensland.com.au/about-epilepsy-epilepsy-queensland/seizure-types/what-are-the-different- types-of-seizures