Intelligent ring system and method based on multi-modal data hierarchical processing

Abstract

The invention discloses an intelligent ring system and method based on multi-modal data layering processing, resource optimization is realized through three-level data layering (a base layer 16B, an original layer 32B and an event layer 8B), event-driven acquisition and cloud cooperative computing are combined, the medical level data quality is ensured, meanwhile, the storage demand is reduced by 57%, the battery life is prolonged by 2.3 times, and the service life of a ring is prolonged by 2.3 times. And the clinical event detection rate is improved by 40%. The system is suitable for long-term monitoring of middle-aged and elderly patients with cardiovascular diseases, sleep, health care and the like, and the practicability and clinical value of wearable equipment are remarkably improved.

Description

1. Technical Field

[0001] The present invention belongs to the technical field of wearable devices, and specifically relates to a smart ring system and method that uses layered data collection and cloud-based collaborative computing. The system is particularly suitable for resource-constrained micro-wearable devices to achieve medical-grade physiological monitoring (such as efficient processing of multi-dimensional data such as cardiovascular parameters, activity status, and environmental adaptation). 2. Background Technology

[0002] Existing smart rings have the following key flaws in data collection and processing:

[0003] 1. The contradiction between computing power and power consumption

[0004] Existing technologies (such as patent CN114224292A) integrate complex algorithms (such as real-time HRV analysis) on the device side, causing the processor to run continuously at high load, power consumption to increase by more than 30%, and battery life to be shortened to 1-2 days, making it difficult to meet long-term monitoring needs.

[0005] 2. Waste of storage resources

[0006] Traditional solutions (such as US20220192521) adopt the "full-time raw data storage" mode, with an average daily storage capacity of 13.12KB. However, only 20% of the data in clinical analysis is related to abnormal events, 90% of the storage space is occupied by non-critical data, and full storage leads to low data retrieval efficiency.

[0007] 3. Lack of clinical data relevance

[0008] Existing systems (such as WO2023115123) do not establish a spatiotemporal correlation between "symptom events and raw data". When users experience abnormalities such as palpitations and falls, they need to manually screen massive amounts of data to trace the source. In clinical applications, the time required for event location increases by 300%, missing the opportunity for intervention. 3. Summary of the Invention

[0009] (1) Purpose of the invention

[0010] To address the problems of "high computing power consumption, storage waste, and weak clinical relevance" in existing technologies, this paper provides a smart ring system and method that combines "data layered processing + event-driven collection + cloud-based collaborative computing". While ensuring medical-grade data quality, it reduces equipment resource consumption and improves the efficiency of clinical event analysis.

[0011] (2) Technical solution

[0012] The present invention achieves optimization through "three-level data hierarchical architecture", "dynamic collection strategy" and "cloud collaborative computing", as follows:

[0013] 1. Three-level data layered architecture

[0014] Divide data into the basic layer, original layer, and event layer to achieve "lightweight storage + precise association":

[0015] The base layer (16B / 5 minutes) is defined as a basic_data_t structure (see Appendix Figure 2 ), including lightweight metrics calculated in real time on the device:

[0016] Core physiological parameters: heart rate (hr), heart rate confidence (hr_confidence);

[0017] ο Activity status: step count (step_count), activity level (activity_level);

[0018] οEnvironment and device status: wearing state (wear_state), temperature compensation value (temp_comp), ambient light intensity (light_level);

[0019] ο Reserved fields: support cloud-based extended computing (such as PTT, HRV, etc.).

[0020] The original layer (32B / event trigger) is defined as the raw_packet_t structure (see the attached Figure 2 ), collect multimodal raw data only during abnormal events:

[0021] ο Synchronization information: synchronization mark (sync_marker), timestamp (timestamp);

[0022] Sensor data: ECG samples (12-bit differential encoding), PPG samples (10-bit), three-axis acceleration (8-bit), temperature / light intensity (4-bit + 4-bit combined encoding);

[0023] o Check information: CRC16 check (crc16), sampling interval (sample_interval).

[0024] The event layer (8B / abnormal trigger) is defined as the event_marker_t structure (see Appendix Figure 2 ), associate events with raw data:

[0025] ο Event ID: event ID (event_id, such as 0x0001 = fall, 0x0002 = abnormal heart rate);

[0026] ο Spatiotemporal information: event time (event_time), severity level (severity);

[0027] ο Related index: original data packet index (related_data).

[0028] 2. Dynamic collection strategy (attached Figure 3 )

[0029] Triggered collection based on device status and abnormal events reduces ineffective consumption:

[0030] 1. Normal monitoring: The device collects basic layer data (16B) at 5-minute intervals and maintains low power consumption mode;

[0031] 2. Anomaly detection: Real-time monitoring of anomalies (e.g., acceleration > 4g, heart rate > 180 beats / minute) through accelerometers and heart rate sensors;

[0032] 3. Trigger acquisition: After an anomaly is detected, 30 seconds of raw layer data acquisition (32B / time) is initiated and an event marker (8B) is generated;

[0033] 4. Data recycling: Raw data during non-event periods is retained for only 2 hours and is overwritten first to save space.

[0034] 3. Cloud Collaborative Computing (Attached Figure 1 )

[0035] The device and the cloud work together to balance computing power and accuracy:

[0036] Light processing on the device: only baseline correction, motion compensation, and simple filtering (such as PPG signal centering) are performed;

[0037] Cloud-based deep analysis: Receives basic data, event markers, and associated raw data to perform HRV analysis, PTT calculation, and multimodal fusion (such as ECG and PPG signal timing alignment).

[0038] (3) Beneficial effects

[0039] 1. Resource optimization: The storage requirement is significantly reduced by 57% (from 13.12KB / day to 5.63KB / day), and the battery life is increased by 2.3 times (from 2 days to 4.6 days), meeting long-term monitoring needs.

[0040] 2. Improved clinical efficiency: Event tagging can directly locate and associate raw data, improving analysis efficiency by 300% and increasing the detection rate of abnormal events such as atrial fibrillation and falls by 40%.

[0041] 3. Strong compatibility supports multi-sensor expansion (ECG / PPG / accelerometer, etc.), adapts to the hardware architecture of smart rings of different brands, and can achieve compatibility through firmware upgrades. IV. Description of the Figures

[0042] Figure 1:System architecture diagram

[0043] · Labeled content: Divided into device layer and cloud layer. The device layer includes "sensor module", "dynamic acquisition module" and "layered storage module", and the cloud layer includes "data analysis engine", "multimodal analysis module" and "event library". Arrows mark the direction of data flow: the device layer transmits basic data, raw data fragments and event markers to the cloud layer via Bluetooth, and the cloud layer returns the analysis results to the device end.

[0044] Description: This demonstrates the overall logic of "device-cloud" collaborative computing and reflects the core design of computing power stratification.

[0045] Figure 2 :Schematic diagram of comparison of three-level data structures

[0046] · Annotated content: The left side shows the original solution data structure (44B original data + no layering), and the right side shows the new solution three-level structure:

[0047] ο Basic layer (16B): lists the 10 fields of basic_data_t and the bytes they occupy;

[0048] Original layer (32B): indicates the compression method of the sensor data (e.g., temperature / light intensity 4-bit + 4-bit merge);

[0049] ο Event layer (8B): indicates that event_marker_t is associated with the index of the original layer (for example, related_data points to the original data packet number).

[0050] Description: This section intuitively presents data simplification and layered design, highlighting optimization points such as "merge encoding" and "differential compression".

[0051] Figure 3 :Flowchart of the method for associating event tags with raw data

[0052] ·Marking content: Process steps:

[0053] a. The device continuously collects basic data (5 minutes / time);

[0054] b. The abnormal detection module is triggered (such as acceleration > 4g);

[0055] c. Start 30 seconds of raw data collection (generating 32B data packets, numbered N1-N5);

[0056] d. Generate event marker (8B), related_data field records N1-N5;

[0057] e. The cloud directly calls N1-N5 raw data for analysis through event marking.

[0058] Description: Demonstrates event-driven collection logic and data association mechanism, reflecting the improvement of clinical traceability efficiency. V. Specific Implementation Methods

[0059] Example 1: Data collection and storage optimization (based on the attached Figure 2 、 3 )

[0060] 1. The basic layer data acquisition device calculates heart rate (hr) at 5-minute intervals through the PPG sensor, accumulates step count (step_count) through the accelerometer, and quantifies ambient light intensity (light_level = 0-100%) through the light sensor. The data is packaged into the basic_data_t structure (16B) and stored in the device flash memory.

[0061] 2. Raw layer trigger conditions When any of the following conditions is met, raw data collection is started:

[0062] οAccelerometer detects∑(a_x 2 +a_y 2 +a_z 2 )>4g and lasting 300ms (fall characteristics);

[0063] Heart rate hr>180 beats / minute or <50 beats / minute, and confidence hr_confidence>80%;

[0064] οThe wearing state wear_state changes from "not worn" to "worn" (triggered by first use).

[0065] 3. When event markers are generated and raw data is collected, event_marker_t is generated synchronously:

[0066] o Fall event: event_id=0x0001, severity=3 (highest level);

[0067] Abnormal heart rate: event_id=0x0002, severity=2;

[0068] The related_data in the tag records the starting index of the original data packet (eg, "10-15" indicates the original data with association numbers 10 to 15).

[0069] Example 2: Collaborative processing between the device and the cloud (based on the attached Figure 1 )

[0070] 1. Device-side preprocessing

[0071]

[0072] 2. Cloud-based in-depth analysis After receiving the raw data, the cloud executes advanced algorithms:

[0073] οHRV calculation: based on ECG R-wave interval (only R-wave is extracted on the device side, SDNN, LF / HF are calculated on the cloud);

[0074] Light intensity compensation: PPG_corrected = RAW_ppg × (1 + 0.05 × (50 - Light_level)), eliminating ambient light interference;

[0075] οEvent tracing: Quickly locate the original data of abnormal periods through event-tagged related_data and generate clinical reports.

[0076] Example 3: Verification of storage optimization effects

[0077] index Original plan Solution of the present invention Optimization range Average daily storage capacity 13.12KB 5.63KB ↓57％ Event location time 120 seconds 40 seconds ↓67％ Battery life 2 days 4.6 days ↑130％ Atrial fibrillation detection rate 65％ 91％ ↑40％

Claims

1. A method for hierarchical processing of smart ring data, characterized in that include: The device continuously generates 16 bytes of basic monitoring data at 5-minute intervals, including heart rate, number of steps, ambient light intensity, and reserved fields; o Response to abnormal events (such as falls, abnormal heart rate) triggers 32B of multimodal raw data acquisition, the raw data including ECG samples (12-bit differential encoding), PPG samples (10-bit), three-axis acceleration (8-bit) and temperature / light intensity (4-bit + 4-bit combined encoding); Generate 8B of event tag data, associating the event type, time, and corresponding original data packet index.

2. The method according to claim 1, characterized in that The abnormal events include: οThe acceleration sensor detects ∑(a_x 2 +a_y 2 +a_z 2 )>4g and last for 300ms; o Heart rate >180 beats / min or <50 beats / min with a confidence level >80%; οThe wearing status switches from "not wearing" to "wearing".

3. A smart ring system, characterized in that The method according to claim 1 or 2, comprising: οDynamic sampling rate control module: collects basic data at 5-minute intervals under normal conditions, and starts 200Hz raw data collection when abnormal; οEvent-driven storage management module: prioritizes retaining event-related raw data, and retains non-event data for only 2 hours; οCloud collaboration interface: The device transmits basic data, event tags and raw data fragments, and the cloud returns in-depth analysis results such as HRV and PTT.

Images