Search

KR-102963783-B1 - SYSTEM FOR PREDICTING RESPIRATORY CONDITION USING ARTIFICIAL INTELLIGENCE AND METHOD THEREOF

KR102963783B1KR 102963783 B1KR102963783 B1KR 102963783B1KR-102963783-B1

Abstract

The present invention relates to a sleep stage prediction system using artificial intelligence and a method thereof. According to an embodiment of the present invention, the apparatus comprises: a motion detection unit installed on a ceiling or wall surface and detecting the movement of a user; a data extraction unit that extracts the user's respiration, heart rate, and spatial data using the user's data measured from the motion detection unit; an artificial intelligence model creation unit that creates one or more artificial intelligence models by learning the user's state data based on the previously stored respiration, heart rate, and spatial data using a Convolutional Neural Network (CNN) and a Recurrent Neural Network (RNN), respectively; and a state extraction unit that extracts the user's state data by inputting the extracted respiration, heart rate, and spatial data into the artificial intelligence model creation unit, and extracts the user's state by applying one or more extracted data to a tree algorithm.

Inventors

  • 정기섭

Assignees

  • (주)비타

Dates

Publication Date
20260512
Application Date
20221213

Claims (14)

  1. Motion detection unit installed on a ceiling or wall that detects user movement; A data extraction unit that extracts the user's respiration, heart rate, and spatial data using the user's data measured from the motion detection unit; An artificial intelligence model creation unit that creates one or more artificial intelligence models by respectively training the user's state data based on the user's respiration, heart rate, and spatial data extracted through the data extraction unit using a CNN (Convolutional Neural Network) and an RNN (Recurrent Neural Network); and It includes a state extraction unit that inputs the extracted respiration, heart rate, and spatial data into one or more artificial intelligence models to extract the user's state data, and applies the extracted data to a tree algorithm to extract the user's state. The above motion detection unit uses radar, and The above data extraction unit is, Set the two largest signals among the acquired signals as chest movement data and abdominal movement data, respectively, and The derivative values of the chest movement data and the abdominal movement data are calculated respectively, and the chest angle value and abdominal angle value are calculated using the calculated derivative values. Respiratory data including a Respiratory Dynamic Instability (RMI) value is extracted using the difference between the above-calculated chest angle value and the above-calculated abdominal angle value, and The above respiration and heart rate data are classified into time series data and frequency spectrogram data, respectively, and the above spatial data are motion data and distance data, and The aforementioned artificial intelligence model production department, Time series data for the above respiration and heart rate data are trained using a CNN, and frequency spectrogram data for the above respiration and heart rate data are trained using an RNN, respectively. The aforementioned artificial intelligence model production department, An artificial intelligence model is constructed using time-series data for respiration and heart rate data trained using a CNN, and the motion data and distance data respectively trained using an RNN. The aforementioned artificial intelligence model production department, An artificial intelligence model is constructed using data trained on time-series data regarding respiration and heart rate data using an RNN, and data trained on the motion data and distance data respectively using an RNN. The above state extraction unit is, Predicting the breathing state of the user by applying weights to one or more artificial intelligence models produced by the artificial intelligence model production unit, and A respiratory state prediction system in which the sum of the weights for each of the above artificial intelligence models is 1.
  2. delete
  3. delete
  4. delete
  5. delete
  6. delete
  7. delete
  8. A detection stage installed on a ceiling or wall that detects user movement; A data extraction step for extracting the user's respiration, heart rate, and spatial data using the user data measured from the motion detection step; An artificial intelligence model creation step for creating one or more artificial intelligence models by respectively training the user's state data based on the user's respiration, heart rate, and spatial data extracted in the above data extraction step using a CNN (Convolutional Neural Network) and an RNN (Recurrent Neural Network); and A state extraction step comprising inputting the extracted respiration, heart rate, and spatial data into one or more artificial intelligence models to extract the user's state data, and applying the extracted one or more data to a tree algorithm to extract the user's state; The above data extraction step is, A step of setting the two largest signals among the acquired signals as chest movement data and abdominal movement data, respectively; A step of calculating the derivative values of the chest movement data and abdominal movement data respectively, and calculating the chest angle value and abdominal angle value using the calculated derivative values; and The method includes the step of extracting respiratory data including a respiratory dynamic instability (RMI) value using the difference between the calculated chest angle value and the abdominal angle value. The above extracted respiration and heart rate data are classified into time series data and frequency spectrogram data, respectively, and the above spatial data is motion data and distance data, and The above artificial intelligence model production stage is, Time series data for the above respiration and heart rate data are trained using a CNN, and frequency spectrogram data for the above respiration and heart rate data are trained using an RNN, respectively. The above artificial intelligence model production stage is, An artificial intelligence model is constructed using time-series data for respiration and heart rate data trained using a CNN, and the motion data and distance data respectively trained using an RNN. The above artificial intelligence model production stage is, An artificial intelligence model is constructed using data trained on time-series data regarding respiration and heart rate data using an RNN, and data trained on the motion data and distance data respectively using an RNN. The above state extraction step is, Predicting the breathing state of the user by applying weights to one or more artificial intelligence models produced by the artificial intelligence model production unit, and A breathing state prediction method in which the sum of the weights for each of the above artificial intelligence models is 1.
  9. delete
  10. delete
  11. delete
  12. delete
  13. delete
  14. delete

Description

System for Predicting Respiratory Condition Using Artificial Intelligence and Method Thereof The present invention relates to a system for predicting breathing status using artificial intelligence and a method thereof. Polysomnography is a test used to diagnose sleep disorders. It comprehensively measures brain waves, eye movements, muscle movements, respiration, and electrocardiograms during sleep, while simultaneously videotaping the sleep state. The recorded data is then analyzed to diagnose sleep-related conditions and determine treatment strategies. The aforementioned polysomnography can diagnose symptoms such as sleep apnea, sleep disorders, and sleepwalking; indices used to assess these conditions include sleep stages, the Apnea-Hypopnea Index (AHI), and Respiratory Effort-Related Arousals (REA). Meanwhile, polysomnography utilizes a manual sleep scoring method in which specialized personnel combine the patient's biometric data measured by various sensors to determine the aforementioned indices. In addition, there is a problem that the time required for sleep scoring becomes too long as it is performed manually by specialized personnel. In fact, it is known that it takes about 3 to 4 hours for a skilled professional to perform sleep scoring for one patient. Consequently, to obtain an accurate diagnosis of sleep disorders, a two-day, one-night examination must be conducted at a sleep center equipped with specialized facilities and equipment staffed by trained professionals. The biosignal monitoring device primarily used to analyze a patient's sleep stages in the diagnosis of sleep-related disorders such as insomnia and narcolepsy is the EEG. In addition, EOG and chin-EMG can be used as auxiliary tools to assess sleep stages. Generally, the procedure for diagnosing sleep disorders, such as insomnia, is as follows. Typically, at a hospital equipped with a polysomnography center, EEG, EOG, and Chin-EMG are measured during sleep for a set period. Based on these measurements, a qualified specialist manually analyzes the sleep stages under the supervision of a sleep specialist to derive results. Based on these results, the sleep specialist diagnoses the sleep disorder, determining the presence of conditions such as insomnia or narcolepsy. In other words, since the determination of sleep stages serves as fundamental data for diagnosing sleep-related disorders, the accuracy of this assessment is a critical factor for diagnosis. However, as previously mentioned, even if sleep stage analysis is performed manually by trained experts, significant errors are inevitable due to the atypicality of brain waves that vary from patient to patient and the complexity of the AASM scoring rules that serve as the criteria for judgment. In particular, disagreements among experts regarding the classification of sleep stages frequently occur due to the presence or absence of judgments based on sleep spindle and K-complexes signals. In addition, human error resulting from the manual interpretation of polysomnography results averaging about eight hours also contributes to the error. The problems with polysomnography include the economic costs associated with using specially trained personnel and facilities, and the presentation of inaccurate experimental results due to environmental changes resulting from sleeping for two days and one night in an unfamiliar environment and with equipment, which leads to the first night effect and inaccurate results regarding sleep changes based on individual physical and psychological conditions. Furthermore, there is the inconvenience of patients having to visit the hospital multiple times to check test results, and there is a problem in that there is a demand from both doctors and patients for the diagnosis of sleep disorders at an affordable price and over a long period of time, tailored to individual needs. Previously, PSGs such as EEG, EOG, and EMG were used to predict sleep stages, calculate sleep stages, measure sleep quality, and identify the causes of sleep disorders and the effects of improvement after treatment; however, wearing sensors for conventional PSG testing degrades the quality of sleep itself, and continuous long-term monitoring is difficult. And although there is an AASM standard for sleep phases, the agreement rate for each sleep article is only 61%, and the accuracy of the radar's HR's RRI (R to R peak Interval) is low. The technology forming the background of the present invention is disclosed in Korean Registered Patent No. 10-23714436 (published March 8, 2022) and Registered Patent No. 10-1235441 (published February 20, 2013). FIG. 1 is a drawing for explaining a breathing state prediction system according to an embodiment of the present invention. FIG. 2 is a diagram illustrating extracted data according to an embodiment of the present invention. FIG. 3 is a diagram illustrating frequency spectrum data according to an embodiment of the present invention. FIG. 4 is a diag