SQNN: a spike-wave index quantification neural network with a pre-labeling algorithm for epileptiform activity...

Area of Science:

• Neuroscience and Neurology
• Artificial Intelligence in Medicine
• Signal Processing

Background:

• Electrical status epilepticus during slow sleep (ESES) is characterized by significant epileptiform activity on electroencephalograms (EEGs) during sleep.
• The spike-wave index (SWI) quantifies this activity in children with ESES, crucial for diagnosis and prognosis.
• Current SWI quantification is often manual, time-consuming, and subjective.

Purpose of the Study:

• To develop and validate a deep learning-based automated method for accurate SWI quantification.
• To improve the efficiency and reliability of SWI measurement in pediatric epilepsy diagnosis.
• To provide a robust tool for clinical decision-making in ESES cases.

Main Methods:

• Development of a pre-labeling algorithm (PreLA) using adaptive wavelet decomposition and a slow-wave discrimination rule for efficient EEG data labeling.
• Construction of a spike-wave index quantification neural network (SQNN) for automated classification of EEG signal points as normal or abnormal.
• Calculation of SWI based on the total duration of identified abnormalities and signal length.

Main Results:

• The PreLA demonstrated effectiveness and robustness in labeling large-scale EEG datasets.
• The SQNN accurately and reliably quantified SWI without requiring predefined thresholds, achieving an average estimation error of 3.12%.
• The automated method was 100 times faster than expert analysis, showing superior accuracy and robustness.

Conclusions:

• Deep learning offers a novel and effective approach for automatic SWI quantification in ESES.
• The PreLA facilitates efficient labeling of EEG data for ESES syndromes.
• The proposed method shows high potential for improving clinical diagnosis and prognosis of epilepsy in children.