Personal tools

Graph neural networks for epileptic seizure detection

From iis-projects

Revision as of 12:24, 12 March 2021 by Xiaywang (talk | contribs)
Jump to: navigation, search
Non-EEG Seizure.jpg

Description

Epilepsy is a severe and prevalent chronic neurological disorder affecting 1–2% of the world’s population [1]. One-third of epilepsy patients continue to suffer from seizures despite best possible pharmacological treatment [2]. For these patients with so-called drug-resistant epilepsy [3], various algorithms based on intracranial electroencephalography (iEEG) recording are proposed to detect the onset of seizures [1]. Complementary to this approach, efficient and robust algorithms are required to not only detect the seizure onset but also to identify the ictogenic (i.e. seizure-generating) brain regions for possible surgical removal [4, 5]. The iEEG currently provides the best spatial resolution and the highest signal-to-noise ratio (SNR) of electrical brain activity recordings [1]. Recent studies have shown successful applications of machine learning methods [1, 6, 7, 8] using iEEG signals to detect two distinct states of brain activity in patients with epilepsy, i.e., interictal (= between seizures) and ictal (= during seizures). These methods are based on extracting useful features followed by traditional supervised machine learning methods (such as random forest [1], support vector machines [6], Bayesian analysis [8], artificial neural networks [6]), and more recently deep learning algorithms [7].

Graph Neural Networks (GNNs), have gained increasing interest in the deep learning(DL) community thanks to their capacity of capturing relational information between entities [9]. Graph theory analysis has been applied to neural signals to analyze the functional connectivity in the human brain [10]. However, conventional GNNs only capture static information, while dynamic graphs take into consideration also the relationship over time between different entities of the graph and their connections. Spatio-temporal graph Convolutional Networks (GCNs) have been proposed by researchers to study the spatial and temporal dependencies in the dataset, for example in traffic forecasting [11]. Continuous-time dynamic graphs have achieved impressive results in many tasks [12]. Few works have applied GNNs to EEG signals, e.g. by combining graphs with convolutional networks (GCNs), achieving state-of-the-art performance in public datasets [13].[14] proposed a temporal GCN to tackle the task of seizure detection. However, EEG signals provide much worse temporal and spatial resolution than iEEG signals.

Depending on the type of thesis, the following steps are to be accomplished:

1 - Development in a high-level programming language (python) of graph neural networks and/or convolutional neural networks for seizure detection.

2 - Benchmarking of these algorithms on a large-scale dataset collected by the Bern Inselspital about epileptic patients (http://ieeg-swez.ethz.ch/).

3 - Comparison with state-of-the-art methods.

4 - Characterization of the algorithm on different computing platforms, from the high-level number of operations to the number of cycles and energy consumption on embedded devices (e.g. GAP8, a multi-core chip from GreenWaves Technology).

Status: Available

Looking for Master's (preferred) or Semester thesis students.

Supervision: Xiaying Wang, Alessio Burrello

Prerequisites

  • Machine Learning
  • Deep Learning
  • Python (and C Programming)


Character

20% Theory
80% Programming

Professor

Luca Benini

↑ top


Literature

  • [1] S. N. Baldassano, B. H. Brinkmann, H. Ung, T. Blevins, E. C. Conrad, K. Leyde,M. J. Cook, A. N. Khambhati, J. B. Wagenaar, G. A. Worrell, and B. Litt, “Crowd-sourcing seizure detection: algorithm development and validation on human im-planted device recordings,”Brain, vol. 140, no. 6, pp. 1680–1691, 2017.
  • [2] D. Schmidt and M. Sillanpää, “Evidence-based review on the natural history of theepilepsies.”Current opinion in neurology, vol. 25 2, pp. 159–63, 2012.
  • [3] J. F. Tellez-Zenteno, R. Dhar, L. Hernandez-Ronquillo, and S. Wiebe, “Long-termoutcomes in epilepsy surgery: antiepileptic drugs, mortality, cognitive and psychoso-cial aspects,”Brain, vol. 130, no. Pt 2, pp. 334–345, Feb 2007.
  • [4] S. Wiebe, W. T. Blume, J. P. Girvin, and M. Eliasziw, “A randomized, controlledtrial of surgery for temporal-lobe epilepsy,”N. Engl. J. Med., vol. 345, no. 5, pp.311–318, Aug 2001.
  • [5] C. Rummel, E. Abela, R. G. Andrzejak, M. Hauf, C. Pollo, M. Muller, C. Weisstan-ner, R. Wiest, and K. Schindler, “Resected Brain Tissue, Seizure Onset Zone andQuantitative EEG Measures: Towards Prediction of Post-Surgical Seizure Control,”PLoS ONE, vol. 10, no. 10, p. e0141023, 2015.
  • [6] A. K. Jaiswal and H. Banka, “Local pattern transformation based feature extractiontechniques for classification of epileptic EEG signals,”Biomedical Signal Processingand Control, vol. 34, pp. 81 – 92, 2017.
  • [7] R. Hussein, H. Palangi, Z. J. Wang, and R. Ward, “Robust detection of epilepticseizures using deep neural networks,” in2018 IEEE International Conference onAcoustics, Speech and Signal Processing (ICASSP), April 2018, pp. 2546–2550.
  • [8] W. Zhou, Y. Liu, Q. Yuan, and X. Li, “Epileptic Seizure Detection Using Lacunar-ity and Bayesian Linear Discriminant Analysis in Intracranial EEG,”IEEE TransBiomed Eng, vol. 60, no. 12, pp. 3375–3381, Dec 2013.
  • [9] J. Zhou, G. Cui, Z. Zhang, C. Yang, Z. Liu, L. Wang, C. Li, and M. Sun, “Graphneural networks: A review of methods and applications,” 2019.
  • [10] S. Sun, X. Li, J. Zhu, Y. Wang, R. La, X. Zhang, L. Wei, and B. Hu, “Graph theoryanalysis of functional connectivity in major depression disorder with high-densityresting state eeg data,”IEEE Transactions on Neural Systems and RehabilitationEngineering, vol. 27, no. 3, pp. 429–439, 2019.
  • [11] B. Yu, H. Yin, and Z. Zhu, “Spatio-temporal graph convolutional networks: Adeep learning framework for traffic forecasting,”Proceedings of the Twenty-SeventhInternational Joint Conference on Artificial Intelligence, Jul 2018. [Online].Available: http://dx.doi.org/10.24963/ijcai.2018/505
  • [12] E. Rossi, B. Chamberlain, F. Frasca, D. Eynard, F. Monti, and M. Bronstein, “Tem-poral graph networks for deep learning on dynamic graphs,” 2020.
  • [13] X. Lun, S. Jia, Y. Hou, Y. Shi, Y. Li, H. Yang, S. Zhang, and J. Lv, “Gcns-net: Agraph convolutional neural network approach for decoding time-resolved eeg motorimagery signals,” 2020.
  • [14] I. Covert, B. Krishnan, I. Najm, J. Zhan, M. Shore, J. Hixson, and M. J. Po,“Temporal graph convolutional networks for automatic seizure detection,” 2019.


Practical Details

↑ top