A new video imaging system for classifying epileptic seizure types

A new 3D video imaging tool shows great potential for real-time classification of epileptic seizures in hospitals. scientific report.

“A seizure is a defining symptom [of epilepsy], and their morphology is of paramount importance for the differential diagnosis and localization of seizure initiation zones in the brain. ” Tomas caracsony, masterA PhD candidate and machine learning researcher at the Institute of Systems Engineering, Computers, Technology and Sciences in Portugal, writes a colleague.

According to researchers, Seizure analysis is currently based on visual interpretation of 2D video-EEG data in epilepsy monitoring units by expert clinicians, with limited semiotic evaluation.

Karácsony et al. presented a novel learning-based approach for motor-based three-class classification of seizures and non-seizure classes in frontal and temporal lobe epilepsy. Their tool uses 3D videos of his seizures acquired with an epilepsy monitoring unit.

The authors implemented “intelligent cropping” by combining Mask R convolutional neural network video cropping and depth cropping-based preprocessing, achieving success rates of 96.52% and 95.65%, respectively. The author then used his 3D depth cropping to remove occlusion and irrelevant information from the scene. This “greatly improved” the classification performance.

From there, a new action recognition approach was used for I3D feature extraction and long short-term memory FC and I3D classification.

“As far as we know, [this approach] We demonstrate that video-based seizure classification outperforms all previous deep learning-based approaches and is likely to support physicians in making diagnostic decisions,” the authors wrote. This study demonstrates the feasibility of our action recognition approach to discriminate between these three classes with samples of only 2 seconds.Furthermore, we evaluated temporal augmentation techniques, resulting in large datasets suggests that we might benefit more from such an extension, but in this case it suffers from a loss of generalization and hence performance.”