Device-based diabetes measurement and risk control device

Abstract

A device for measuring and monitoring diabetes, consisting of a system-on-chip for real-time processing of sensor data, several sensors connected to the system-on-chip, including a blood glucose sensor, a heart rate sensor and a blood pressure sensor, a machine learning model implemented on the system-on-chip for analyzing the sensor data and generating a diabetes risk score, and a memory buffer within the system-on-chip for storing sensor data for trend analysis and improved prediction accuracy.

Description

AREA OF INVENTION

[0001] The invention relates to IoT health monitoring devices with integrated biosensors and a system-on-chip that runs machine learning models on the device to estimate glucose trends and diabetes risk in real time. BACKGROUND OF THE INVENTION

[0002] Diabetes management benefits from continuous monitoring of glucose and vital signs. However, conventional methods rely on invasive blood glucose measurements via finger prick or minimally invasive CGM systems, which are problematic in terms of cost, calibration, and comfort. Fully non-invasive sensors remain the subject of intensive research, with challenges to performance and reliability during movement and under varying physiological conditions. IoT-enabled wearables and home devices can stream biosignals for analysis, while machine learning improves the prediction of hyper- / hypoglycemia and disease risk using multimodal input data such as continuously monitored glucose levels, heart rate, and blood pressure. Decision trees and ensemble models have been investigated for diabetes prediction in IoT environments.They often combine physiological and behavioral data to calculate risk scores and trigger timely interventions. Evidence suggests that inference directly on the device or at the network edge can reduce latency and connectivity dependency. A device-centric design that integrates sensors and machine learning inference on a system-on-chip with a local memory buffer for trend analysis can enable robust, private, and responsive monitoring suitable for everyday use and integration into broader treatment pathways. SUMMARY OF THE INVENTION

[0003] The invention relates to a device for measuring and monitoring diabetes. This device comprises a system-on-chip (SoC) for real-time processing of sensor data, several IoT-based sensors connected to the SoC (including at least one blood glucose sensor, one heart rate sensor, and one blood pressure sensor), a machine learning model (e.g., a decision tree) implemented on the SoC, which has been trained to analyze real-time data and generate a diabetes risk score, and a memory buffer in the SoC for storing sensor data for trend analysis and improved predictive accuracy. The device performs preprocessing, feature extraction, and inference directly on the device to estimate the current status and short-term risk.Optionally, summaries for long-term analysis can be synchronized with cloud services or mobile applications, while the core functionality is retained offline through local storage and processing. DETAILED DESCRIPTION

[0004] The device integrates a glucose measurement method, which can be a minimally invasive CGM system or a non-invasive sensor, as well as heart rate and blood pressure sensors. The SoC collects synchronized readings and applies filtering and calibration to compensate for motion and physiological fluctuations described in wearable monitoring studies. A real-time pipeline on the SoC derives features such as the glucose change rate, heart rate variability, and blood pressure trends, and then uses a trained decision tree classifier or regressor to calculate a diabetes risk score and event probabilities. This is analogous to machine learning frameworks that combine physiological data streams for diabetes prediction in an IoT context. The SoC has a memory buffer that manages sliding time windows (e.g., hours to weeks) for trend analysis and personalized baselines.These features are associated with improved predictive accuracy in the literature for wearable glucose prediction and anomaly detection. A communication stack enables optional synchronization with a mobile app or cloud platform for visualization and clinical evaluation. Simultaneously, it ensures on-device data processing to guarantee low latency and reliability in the event of connection interruptions during daily use. The system implements model update mechanisms that securely provide new decision tree parameters or supplementary ensemble models. This allows the system to be adapted to individual patterns without affecting local operation and while respecting data privacy preferences. Power management optimizes sampling and evaluation schedules to balance battery life and monitoring accuracy.The device can reduce sensor activity and compress stored data while preserving key features for later analysis. To address the known limitations of non-invasive glucose monitoring, the design supports sensor quality assessment and, where available, utilizes validated minimally invasive inputs. This aligns with findings from studies highlighting the precision limitations and calibration requirements of newer non-invasive methods. Safety logic limits predictions and provides user alerts if sensor quality deteriorates or risk values ​​exceed thresholds. This enables timely self-management or referral to healthcare teams, as recommended in IoT-based diabetes monitoring systems.The housing and user interface provide clear indicators for sensor and risk status, while APIs offer summary metrics for integration into electronic health records or care platforms within diabetes treatment programs. The device's form and functional requirements comply with German utility model practice, which protects product claims through rapid registration following formal examination, thus enabling early enforceability while simultaneously allowing for further model development.

Claims

[1] A device for measuring and monitoring diabetes, consisting of a system-on-chip for real-time processing of sensor data, several sensors connected to the system-on-chip including a blood glucose sensor, a heart rate sensor and a blood pressure sensor, a machine learning model implemented on the system-on-chip for analyzing the sensor data and generating a diabetes risk score, and a memory buffer within the system-on-chip for storing sensor data for trend analysis and improved prediction accuracy. [2] Device according to claim 1, wherein the machine learning model comprises a decision tree trained on multimodal physiological data to classify diabetes risk states and predict short-term glycemic events based on derived features. [3] Device according to claim 1 or 2, which is configured to manage the memory buffer as sliding windows of synchronized sensor measurements to support personalized baselines and model fitting, and wherein the device performs preprocessing, calibration and feature extraction on the device to reduce latency and dependence on connectivity. [4] Device according to any of the preceding claims, wherein the glucose measurement comprises minimally invasive continuous glucose monitoring with quality controls and calibration, and the device optionally synchronizes summary data with external applications while maintaining core operation offline.