Seizure detection and epileptogenic zone localisation on heavily skewed MEG data using RUSBoost machine learning...

Area of Science:

• Neuroscience
• Medical Technology
• Signal Processing

Background:

• Epilepsy is a neurological disorder characterized by recurrent seizures.
• Magnetoencephalography (MEG) is crucial for pre-surgical localization of the epileptogenic zone.
• Manual detection of seizures in MEG data is time-consuming and laborious.

Purpose of the Study:

• To develop a novel, automated method for detecting ictal activities in MEG.
• To improve the efficiency and accuracy of seizure onset zone localization.
• To reduce the manual workload in MEG data analysis for epilepsy patients.

Main Methods:

• Extraction of statistical features: short-time permutation entropy (STPE), gradient of STPE (GSTPE), short-time energy (STE), and short-time mean (STM).
• Classification of ictal and interictal MEG data using the RUSBoost algorithm with k-fold cross-validation.
• Region-specific classification for epileptogenic lobe localization.

Main Results:

• Achieved 93.4% accuracy for seizure detection.
• Obtained 93% specificity and sensitivity for seizure detection.
• Localization accuracies for each lobe ranged from 88.1% to 99.1%.

Conclusions:

• The proposed automated method significantly reduces the laborious manual inspection of ictal MEG data.
• This system offers a more efficient and accurate real-time solution for seizure detection and localization.
• Provides valuable insights for clinicians in identifying the epileptogenic lobe and guiding source localization.