This AI Paper Presents A Novel 3D Video Action Recognition Deep Learning Approach For Near Real-Time Epileptic Seizure Classifcation

A significant portion of the world’s population suffers from epilepsy, a chronic neurological disorder that is relatively frequent. The primary symptom is a seizure, and the kind of seizure (semiology) is crucial for differentiating diagnoses and pinpointing the location of the seizure start zone in the brain. This is crucial for individuals with drug-resistant epilepsy who are being considered for epilepsy surgery. Seizure analysis currently relies on highly specialized clinicians’ visual interpretation of 2D video-EEG data in epilepsy monitoring units (EMUs), where semiology evaluation is constrained by high inter-rater variability.

Quantitative seizure classification studies are relatively uncommon, despite the abundance of video content that is readily available. Though promising in the research, automated and semi-automatic computer-vision analysis systems still need a significant amount of “human in the loop” work. Approaches for automated, 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 biggest 3D-video-EEG collection in the world, the Neurokinect 3D video dataset, we processed IR seizure films using Inception-V3 feature extraction and a fully connected classifier, yielding a moderate result (AUC 0.65).

Table 1: Comparison of the results

They contend that the classifier’s object identification training needed temporal information and may have caused it to be biased toward one class and provide mediocre results. Other research processed three primary parallel threads of body areas and posture using a hierarchical technique. The “leave one subject out” cross-validation produced modest accuracy (50.9-69.8%), indicating the inability to capture subject-invariant features and subsequent overflight to subject-specific facial features and posture coordinates. Accuracy was high when training and validation used the same subjects. A shallow CNN and LSTM-based architecture were also utilized in the literature, but no significant improvement was seen (62.2–66.5%).

👉 Read our latest Newsletter: Microsoft’s FLAME for spreadsheets; Dreamix creates and edit video from image and text prompts......

The authors present a novel contribution that was motivated by the way epileptologists analyze seizure semiology, where they consider not only the presence of particular “Movements Of Interest” (MOI) in various body regions of the patients but also their dynamics (the order in which they appear) and biomechanics characteristics (such as speed/acceleration patterns, movement amplitude, etc.). They, therefore, decided to research the viability of a 3-class general, cross-subject, near-real-time epileptic seizure classification pipeline for 24/7 automated seizure detection at the EMUs to explore the incorporation of these Spatio-temporal elements.


Check out the Paper. All Credit For This Research Goes To Researchers on This Project. Also, don’t forget to join our Reddit page and discord channel, where we share the latest AI research news, cool AI projects, and more.

Aneesh Tickoo is a consulting intern at MarktechPost. He is currently pursuing his undergraduate degree in Data Science and Artificial Intelligence from the 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 loves to connect with people and collaborate on interesting projects.