A personalized method for predicting epilepsy crises before / at the time of the crisis

Description

[0001] A PERSONALIZED METHOD FOR PREDICTING EPILEPSY CRISES BEFORE / AT THE TIME OF THE CRISIS

[0002] Technical Field

[0003] The invention relates to a method for detecting epileptic seizures of epilepsy patients together with their type (epileptic / non-epileptic) at the time of the seizure and / or detecting them before the seizure.

[0004] Prior Art

[0005] It has been observed that various studies have been carried out in the field of predicting epilepsy crises in the art.

[0006] In the International Patent document numbered WO2022104412A1, which is in the known state of the art, methods and systems developed for determining the therapeutic window, detection, prediction, classification of neuroelectrical, cardiac and / or pulmonary events and optimization of treatment accordingly are mentioned. The document mentions the training of sensor data using artificial intelligence.

[0007] In the Korean Patent document numbered KR102298943B1, which is in the known state of the art, an artificial intelligence-based information method that provides information for brain information and brain activation determined while the user is in motion is mentioned.

[0008] In the UK Patent document numbered GB2588272A, which is in the known state of the art, a wearable, portable brain function monitoring device with a headband and a muscle ring (figure 5, 6), which can be worn on the arm or leg, in which a plurality of graphene and / or photonic sensors are embedded in the fabric, is mentioned.

[0009] In the Australian Patent document numbered AU2021101132A4, which is in the known state of the art, the analysis and monitoring of brain waves for the treatment of epilepsy using artificial intelligence is mentioned.

[0010] When the existing studies in the art are examined, successful results cannot be obtained in every patient because the studies are based on general learning. This is because each person is different from the other. At the time of crisis, the patient's condition and location cannot be communicated to relatives. Since the decision-making process is based on a microcontroller, a high-capacity processor is used in the device and therefore requires a battery of higher capacity and size. This increases the size of the device. Patient data cannot be recorded in a way that can also be transmitted to the doctor.

[0011] When the existing works in the art are examined, the subject of the invention is the need to develop a method that enables the detection of epilepsy seizures of epilepsy patients with the type of epilepsy crisis at the time of the crisis and / or detection before the crisis.

[0012] Objectives of the Invention

[0013] The object of the present invention is to provide a method for detecting epileptic seizures of epilepsy patients at the time of the seizure and / or before the seizure.

[0014] Another object of the present invention is to provide a method for detecting the type of epileptic seizure.

[0015] Detailed Description of the Invention

[0016] The invention relates to a method for detecting epileptic seizures of epilepsy patients at the time of the seizure, together with the type of seizure (epileptic or non-epileptic) and / or before the seizure, and compromises the following steps:

[0017] - realization of mobile device pairing with a wristband / armband containing a wearable sensor,

[0018] - reading sensor data via the sensor paired with the mobile device,

[0019] - sending the sensor data to the mobile device via bluetooth if the rate of change in sensor data is more than 5%,

[0020] - also periodically sending the battery status data to the device,

[0021] - wearable sensor data received in time-frequency planes to extract statistical and subband features,

[0022] - classification of the extracted statistical and sub-band features by deep learning, artificial neural networks, support vector machines, decision trees, etc. with classifier artificial intelligence methods,

[0023] - as a result of the classification, first of all, determining whether there is a crisis or not, if there is a crisis, its type, if not, whether it is a priori crisis or not, - retraining module from the analyzed data and detecting the data marked as erroneous by the user and sending it to the server,

[0024] - determination of the ideal features and parameters of artificial intelligence methods obtained by classifier artificial intelligence methods such as deep learning, artificial neural networks, support vector machines, decision trees, etc. on the server,

[0025] - sending the artificial intelligence method to the mobile software with the newly determined ideal features and parameters for the user as a result of the training and making decisions after the training with the newly selected features and method,

[0026] - finally, notification of an epilepsy crisis to the user at the time of the crisis and / or before the crisis.

[0027] In the inventive method, a wristband / armband containing a wearable sensor is paired with a mobile device. In this way, data security is ensured. Sensor data is read in the paired sensor node. If the rate of change in sensor data is less than 5%, sensor data is not sent to the mobile device. Data is sent to the mobile device via bluetooth. At the same time, data showing the battery status is periodically sent to the mobile device.

[0028] The received wearable sensor data is analyzed in a time-shifted manner. First, features are extracted from the sensor data with signal processing methods. The extracted features are analyzed with artificial intelligence. As a result of the analysis, the following operations are performed. i. Crisis detection ii. Determination of the type of crisis if there is a crisis

[0029] (1) Epileptic crisis

[0030] (2) Non-epileptic crisis iii. If there is no crisis, determining whether there is an a priori crisis

[0031] The method uses 3 different classifiers for classification purposes. These classifiers are used to determine whether there is a crisis, crisis type and a priori crisis warning, respectively.[1]

[0032] For the classification process, the ideal features determined for each classifier and the signals received from wearable sensors are used as input to the ideal classifier.

[0033] In the classification process, features are extracted from sensor data received from wearable sensors and the extracted features are processed in the classifier and a decision is formed as a result. In the inventive method, false detections are recorded with time. When the user runs the retraining module in the software, the following operations are performed. i. Incorrect data is sent to the server. ii. The server determines the ideal personalized features and parameters as well as the artificial intelligence method. iii. As a result of the training, the artificial intelligence method is sent to the mobile device with the newly determined ideal features and parameters. iv. Decisions to be taken after the training are made according to the new features and artificial intelligence method.

[0034] The user can either revert to the features and Al method selected before the last training or use the initial features and Al method.

[0035] On the server, the ideal feature selection (after the ideal classifier is determined, the features that achieve the highest success with the minimum number of features are determined by testing a set of possible features and then only those are used) and the parameters of the artificial intelligence method (Each method has its own specific parameters. For example, in neural networks, the ideal number of layers, the number of neurons in each layer and the transfer function are determined.) are indicated.

[0036] The inventive method also has the following features. i. Saving crisis information on the server ii. Sharing of the stored data with the persons requested / approved by the user and / or his / her doctor iii. When the crisis information of the user is generated, the crisis information and the location of the user (location) are received via GPS and, if any, sent to the people defined by the user via SMS and e-mail iv. The medication(s) defined by the user or the user's doctor, if defined by the user, are recorded on the server and mobile device along with the time of use. After registration, the medication is tracked and the user is warned every 5 minutes if the medication is not used within the defined time interval; if the medication is not used within 30 minutes, the user's relatives, if defined by the user, and the user's doctor, if defined, are warned via SMS and e-mail. In addition, in the method of the invention, each user's seizure information and payment information are stored on the server consisting of a database and a web server.

[0037] In the method of the invention, a continuous learning feature, which can be referred to as personalized learning, has been added to the device to overcome the general learning problem. In the event of a crisis, the ability to communicate the patient's crisis information and location to relatives has been added. The decision-making process was transferred from the microcontroller to the mobile device. In this way, the size of the device has been reduced as well as the battery used in the device where the method is used. A server was integrated into the system for continuous recording of the data.

[0038] References:

[0039] [1] Ethem Alpaydin. Introduction to Machine Learning (Adaptive Computation and Machine Learning Series). The MIT press, 2004. ISBN: 026201211 1 Price £32.95 / $50.00 (hardcover), xxx+415 pages.

Claims

CLAIMS1. The invention relates to a method for detecting epileptic seizures of epilepsy patients at the time of the seizure together with the type of seizure (epileptic or non-epileptic) and / or before the seizure, compromises the following steps,- realizing mobile device pairing with a wristband / armband containing a wearable sensor,- reading sensor data from the sensor paired with the mobile device,- sending the sensor data to the mobile device via bluetooth if the rate of change in sensor data is more than 5%,- as well as periodically sending battery status data to the device,- extraction of statistical and sub-band features by converting the received wearable sensor data into time-frequency planes,- Classification of the extracted statistical and sub-band features by deep learning, artificial neural networks, support vector machines, decision trees, etc. classifier artificial intelligence methods,- determining whether there is a crisis as a result of the classification, if there is a crisis, the type of crisis, if not, whether it is a priori crisis,- retraining module from the analyzed data, detecting the data marked as erroneous by the user and sending it to the server,- Determination of ideal features and parameters of artificial intelligence methods obtained by classifier artificial intelligence methods such as deep learning, artificial neural networks, support vector machines, decision trees, etc. on the server,- sending the artificial intelligence method to the mobile software with the newly determined ideal features and parameters for the user as a result of the training and making decisions after the training with the newly selected features and method,- finally, notification of an epilepsy crisis to the user at the time of the crisis and / or before the crisis2. The invention relates to a method as in claim 1, is characterized by the fact that it compromises the steps of,- Saving crisis information on the server,- Receiving crisis information and the location of the user via GPS when crisis information is generated for the user, and communicating it via SMS and e-mail to the people defined by the user, if any,

Images