Glucose monitoring method and system
Inventors
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-12263021-B2
Publication Date
2025-04-01
Expiration Date
Abstract
This invention provides a system and method to protect an artificial pancreas' sensor, infusion system, and alert systems from EMI/wireless attacks using a medical software or application, close the gap between sensor glucose and blood glucose, and build a non-invasive hypoglycemia and hyperglycemia false alarm detection scheme with the help of a wristband. This inventive method and system provides a more accurate blood glucose prediction. It comprises preprocessing the CGM readings with Kalman smoothing for sensor error correction improves the robustness of the BG prediction. In one or more embodiments, the inventive system and method uses one or more physiological information such as meal, insulin, aggregations of step count, and preprocessed CGM data. The invention provides a novel approach for leveraging the stacked LSTM based deep RNN model to improve the BG prediction accuracy. The invention provides a special circuits-Transduction Shield- to detect and correct the sensor errors caused by EMI attacks.
Core Innovation
The invention relates to a diabetes monitoring system that determines a false alarm in a continuous glucose monitor by continuously monitoring measured patient glucose levels together with multiple patient physiological signals. The patient physiological signal detector continuously monitors a plurality of patient physiological signals selected from heart rate, galvanic skin response, skin tremor, and heart rate variability and outputs the monitored physiological signals to a receiver and processor.
The processor continuously applies a stacked LSTM based deep recurrent neural network to the patient physiological signals and the measured patient glucose levels to predict a patient's predicted blood glucose level. The processor compares the predicted blood glucose level to the measured patient glucose level to calculate a confidence score, and actuates an alarm when the confidence score is less than a predetermined threshold level, to suppress false alarm conditions in a continuous glucose monitoring context.
The invention also addresses false hypoglycemia and hyperglycemia alarms by incorporating additional physiological and activity signals and by preprocessing continuous glucose monitoring signals using Kalman smoothing. It further includes a hardware/software transduction shield circuit to capture and correct EMI-corrupted sensor signals, with sensor anomaly detection and correction based on transduction-aware sensitivity to EMI.
Overall, the invention combines continuous physiological monitoring, deep recurrent neural network prediction with a confidence-score comparison, Kalman-smoothed CGM preprocessing, and EMI-robust sensor correction/transduction shielding to improve the determination of false alarms within a continuous glucose monitor for an artificial pancreas configuration.
Claims Coverage
The independent claim for a system for determining a false alarm in a continuous glucose monitor includes three core inventive elements: continuous patient physiological signal monitoring, stacked LSTM deep recurrent neural network prediction with confidence-score calculation, and alarm actuation based on the confidence score relative to a predetermined threshold level.
Continuously monitoring patient physiological signals for false-alarm determination
A patient physiological signal detector continuously monitors a plurality of patient physiological signals selected from heart rate, galvanic skin response, skin tremor, and heart rate variability, and outputs the monitored patient physiological signals to the receiver and processor.
Stacked LSTM deep recurrent neural network prediction using physiological and measured glucose
A processor continuously applies a stacked LSTM based deep recurrent neural network to the patient physiological signals and the measured patient glucose levels to predict a patient's predicted blood glucose level and compares the predicted blood glucose level to the measured patient glucose level to calculate a confidence score.
Confidence-score threshold alarm actuation for false alarm suppression
The processor actuates an alarm when the confidence score is less than a predetermined threshold level, while the receiver continuously receives and outputs to the processor the patient physiological signals and the measured patient glucose levels.
Kalman smoothing of measured patient glucose levels
The measured patient glucose levels are preprocessed with Kalman smoothing.
Activity-based verification of the confidence score
An activity detector measures activity data indicating the patient's sleep, exercise, eating, and drinking, and the processor uses this activity data to verify the confidence score.
Wearable physiological signal detector configuration
The patient physiological signal detector is a wearable.
Personal digital assistant receiver configuration
The receiver comprises a personal digital assistant.
Smart phone receiver configuration
The personal digital assistant further comprises the processor and is a smart phone.
Across the independent claim and its dependent refinements, the inventive coverage centers on using continuous patient physiological signals together with measured glucose levels as inputs to a stacked LSTM based deep recurrent neural network, computing a confidence score from the prediction-versus-measurement comparison, and actuating a false-alarm decision when the confidence score falls below a predetermined threshold. Dependent claims further specify Kalman smoothing of glucose levels, wearable detection, personal-digital-assistant and smart-phone receiver configurations, and activity-based verification of the confidence score.
Stated Advantages
Determines a false alarm in a continuous glucose monitor using confidence-score-based alarm actuation.
Suppresses false hypoglycemia and hyperglycemia alarms by using physiological signals and confidence-score logic.
Improves prediction accuracy of blood glucose compared with raw CGM and prior art, as supported by reported RMSE and MAE performance results.
Provides hardware/software transduction shielding and sensor anomaly detection/correction to address EMI-corrupted sensor signals.
Documented Applications
Artificial pancreas diabetes monitoring using a continuous glucose monitor with an insulin pump that injects insulin based on measured glucose levels.
False alarm determination in a continuous glucose monitor using patient physiological signals and alarm actuation based on a confidence-score threshold.
Sensor EMI manipulation experimentation for continuous glucose monitoring and sensor correction validation.
Model and signal-processing evaluation using an OhioT1DM dataset for blood-glucose prediction accuracy comparisons.
Interested in licensing this patent?