METHODS FOR DIAGNOSING PARKINSON'S DISEASE BASED ON MULTIMODAL DATA ANALYSIS USING MACHINE LEARNING (VERSIONS) Russian patent published in 2024 - IPC A61B5/103 G16H50/20 

FIELD: medicine.

SUBSTANCE: group of inventions relates to medical information and communication technologies (ICT), specifically intended for medical diagnostics, and can be used for diagnosing a neurological disease from a set of desired neurological diseases: Parkinson's disease, essential tremor, diffuse Lewy body disease. During the task, the patient collects and records simultaneously video data containing the type of motor activity of the patient, and data from patient-carried sensors—accelerometer, gyroscope and magnetometer. File with key points of the patient's body is obtained for each frame of video data. Patient’s body segments are calculated for each frame based on the obtained key points of the patient’s body. Relative speed and acceleration are calculated based on the calculated segments of the patient's body, and a data set is obtained. Data obtained at the previous stage are segmented into time intervals, features are extracted from the segmented set of video data and corresponding vectors of video data features are obtained. Data obtained using wearable sensors is segmented into time intervals corresponding to segmented video data, features are selected from a segmented data set obtained using wearable sensors, and corresponding vectors of features of wearable sensors are obtained. Combining the video data and wearable sensor feature vectors, which correspond to the same time intervals from the video data and wearable sensor data, into one combined feature vector, and a set of combined feature vectors is obtained. Obtained set of combined feature vectors is classified using machine learning methods and the probabilities of the presence of a disease in the patient from the set of the desired neurological diseases are obtained: Parkinson's disease, essential tremor, diffuse Lewy body disease. Diagnosing the presence or absence of a neurological disease from the set of the desired neurological diseases is performed: Parkinson's disease, essential tremor, diffuse Lewy body disease, using the derived probabilities of the presence of the desired neurological diseases in the patient.

EFFECT: group of inventions provides higher accuracy of diagnosing the desired neurological disease by analyzing multimodal data, including video data containing motor activity of the patient and data from wearable sensors, collected simultaneously during the exercise by the patient.

14 cl, 3 dwg