A significant portion of the world’s population suffers from epilepsy, a relatively common chronic neurological disorder. The primary symptom is a seizure, and the type of seizure (semiology) is crucial in differentiating diagnoses and locating the location of the seizure onset area in the brain. This is crucial for people with drug-resistant epilepsy who are considering epilepsy surgery. Seizure analysis currently relies on visual interpretation by highly specialized clinicians of 2D video-EEG data in Epilepsy Monitoring Units (EMUs), where semiological assessment is limited by high inter-rater variability. .
Quantitative seizure classification studies are relatively rare, despite the abundance of readily available video content. Although promising in research, automated and semi-automated computer vision analysis systems still require a significant amount of “human-in-the-loop” labor. Approaches for automated and AI-supported solutions are much rarer (Table 1). They suggested using IR and depth video data to classify epilepsy using convolutional neural networks (CNNs). To the best of their knowledge, the world’s largest collection of 3D video-EEG, the Neurokinect 3D video dataset, we processed IR capture movies using Inception-V3 feature extraction and d ‘a fully connected classifier, giving a moderate result (AUC 0.65).
They argue that the classifier’s object identification training needed temporal information and may have caused it to be biased towards one class and provide poor results. Other research has addressed three main parallel threads of body and posture areas using a hierarchical technique. Cross-validation “leave a subject out” yielded modest accuracy (50.9-69.8%), indicating inability to capture invariant subject features and subsequent skimming of specific facial features and posture coordinates about. Accuracy was high when training and validation used the same subjects. A shallow architecture based on CNN and LSTM has also been used in the literature, but no significant improvement was observed (62.2–66.5%).
The authors present a new contribution motivated by the way in which epileptologists analyze the semiology of seizures, where they consider not only the presence of particular “Movements of Interest” (MOI) in various regions of the body of patients, but also their dynamics ( the order in which they appear) and biomechanical characteristics (such as velocity/acceleration patterns, range of motion, etc.). They therefore decided to investigate the viability of a 3-class general, between-subjects, and near-real-time seizure classification pipeline for automated 24/7 seizure detection in EMUs. in order to explore the incorporation of these spatio-temporal elements.
Check Paper. All credit for this research goes to the researchers on this project. Also don’t forget to register. our Reddit page and discord channelwhere we share the latest AI research news, cool AI projects, and more.
Aneesh Tickoo is an intern consultant at MarktechPost. He is currently pursuing his undergraduate studies in Data Science and Artificial Intelligence at Indian Institute of Technology (IIT), Bhilai. He spends most of his time working on projects aimed at harnessing the power of machine learning. His research interest is image processing and is passionate about building solutions around it. He enjoys connecting with people and collaborating on interesting projects.
#paper #presents #video #action #recognition #deep #learning #approach #realtime #epileptic #seizure #classification