Epilepsy is one of the most common neurological disorders, affecting 50 million people globally. Epileptic seizures result from abnormal brain activity and can lead to loss of consciousness, uncontrollable movements, and cognitive impairments. Currently, around 70% of epilepsy patients experience seizure control through medical therapy or surgery. Accurate diagnosis and appropriate treatment rely heavily on the identification of epileptic signals in EEG (electroencephalogram) recordings. However, this process is time-consuming, as EEG recordings can span hours or even days. Additionally, distinguishing epileptic signals from other brain activities requires considerable clinical expertise.
To address these challenges, a team of scientists, including researchers from Immanuel Kant Baltic Federal University in Kaliningrad, Russia, has developed a new automated method for detecting epileptic brain activity in EEG recordings. Their approach integrates two techniques: an unsupervised classifier and a trainable neural network. This two-stage detection system simplifies EEG analysis and enhances epilepsy detection accuracy.
In the first stage, the classifier identifies «emissions,» signals that exceed normal brain activity. These emissions could represent seizures, external noise, or other atypical brain activities such as sleep spindles. The classifier flags potential epileptic episodes but also includes false positives. In the second stage, a more complex convolutional neural network, commonly used in image analysis, examines the flagged EEG data. By analyzing the EEG as a whole, the network mimics the way a doctor would assess the signals for seizure markers, improving the accuracy of detection.
Tests using EEG data from 83 epilepsy patients showed promising results. While the sensitivity (ability to detect abnormal signals) of the classifier and neural network alone reached 90% and 96%, their specificity (ability to differentiate epileptic from non-epileptic signals) was lower, at 12% and 13%. However, the combined two-stage system, while slightly less sensitive at 84%, had a much higher specificity of 57%, significantly reducing false positives.
This automated EEG analysis system has the potential to greatly reduce the workload of epileptologists, making the identification of epileptic seizures in long EEG recordings faster and more efficient. According to Alexander Hramov, project lead and researcher, this advancement promises the creation of automated tools that could streamline the work of clinicians, improving care for epilepsy patients.
Article written by Hospi Medica
01/10/2024
Source:
Hospi Medica
https://www.hospimedica.es/cuidados-criticos/articles/294802590/algoritmo-de-ia-automatiza-el-analisis-de-eeg-para-detectar-epilepsia-con-alta-precision.html