Classifying Seizures With The Help Of AI Is Now Possible

Classifying Seizures With The Help Of AI Is Now Possible
SHARE

Using 3D movies and infrared radar, a study team from INESC TEC and the University of Munich, together with Tamás Karácsony, a Ph.D. student at Carnegie Mellon Portugal (CMU Portugal), evaluated an innovative way to classify seizures, which are the most prominent sign of epilepsy.

The results of this study, which were originally posted in Nature Scientific Reports and were managed by Joao Paulo Cunha, who is a researcher at INESC TEC and a professor at FEUP, were presented there.

Even though there is a wide variety of video resources accessible on seizure classification, research on the topic are still uncommon, and techniques for programmed, AI-supported approaches are even less common.

This new study demonstrates a revolutionary technique that is the first to include near-real-time classification from two-second samples. As a result, it presents the possibility of a system that uses deep learning to help diagnostic and monitoring processes. This method makes it possible to differentiate between seizures originating in the frontal and temporal lobes, which are the two most frequent types of epilepsy, and occurrences that are not associated with epilepsy.

Epilepsy is a chronic neurological disorder that affects one percent of the world’s population. Seizures are one of the most prominent signs of epilepsy, and the study of seizure semiology is critical for accurately diagnosing probable episodes. The evaluation of a seizure is often carried out in epilepsy monitoring units, also known as EMUs, by trained medical experts making use of 2D video-EEG technology.

The semiology evaluation, on the other hand, is hampered by a high level of inter-rater variability among the aforementioned specialists. Furthermore, despite their promise, the automated and semi-automatic approaches utilizing computer vision are still dependent on a significant amount of “human in the loop” labor.

In most cases, a patient is observed for many days, and then, subsequently, they must have a comprehensive examination for the seizures. The clinical personnel must devote a significant amount of their time and effort to this endeavor. In order to circumvent this issue, the research group has developed a system based on deep learning that is capable of the near real-time and automated classification of epileptic episodes.

The study team accessed the biggest 3D video-EEG collection in the world and obtained films of 115 seizures. They then developed a semi-specialized and programmable pre-processing method to remove undesired surroundings from the movies.

Two picture cropping approaches are merged in realistic terms: depth and Mask R-CNN. This offers a clean scenario while also refining the retrieval of pertinent information from the movies that are accessible, eliminating variations that are not linked to the seizure, and improving the process of identifying seizures from other types of motion.

This solution removes information that is not relevant to the problem at hand, such as physicians moving about patients, by using a method based on action detection and cropping the scene in an intelligent manner in three dimensions. This work has also shown the applicability of the action-recognition system to classify two classes of epilepsy as well as the class of non-epileptic patients using only two seconds of data, making it appropriate for near-real-time monitoring of patients. In addition, the technique that has been described may be used in various 3D video datasets for the purpose of monitoring and analyzing seizures.

It is possible to utilize it for monitoring and alerts, which may alert personnel; or, if the method is moved to an ambulatory environment, a caregiver, while a seizure is continuing, resulting in a quicker reaction, which would reduce the related dangers and the risk of sudden unexpected death in epilepsy (SUDEP). Additional research must be conducted before the clinical use of this technology may begin. On the other hand, it is anticipated that the system will, in the long term, be beneficial for the patients, the physicians, and the clinics. With the help of automated diagnostic assistance, physicians need to spend less time studying the films. As a result, they are able to treat more patients and, with any luck, come to more informed judgments, which in turn lowers the expenses connected with running clinics.


SHARE

Anna is an avid blogger with an educational background in medicine and mental health. She is a generalist with many other interests including nutrition, women's health, astronomy and photography. In her free time from work and writing, Anna enjoys nature walks, reading, and listening to jazz and classical music.

Post Comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.