Wearable Sensor Data Labeling Using Medical Evaluation Results
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Wearable IoT devices, such as smart watches and patches, provide inaccurate health parameter determinations due to insufficient training data, which can be improved by automatically labeling sensor data with medical evaluation results during facility visits.
Innovation Solution
A system that identifies when a user is at a medical facility, monitors sensor data during evaluations, retrieves medical evaluation results, and labels the data accordingly, using GPS, accelerometer data, and user interaction to enhance training data accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If wearable IoT devices use consumer-grade sensors for health measurements, then device accessibility and ease of use are improved, but measurement precision and reliability deteriorate
Solution Approach 1:
The patent uses medical facilities and medical-grade sensors as an intermediary to bridge the gap between consumer-grade sensors and accurate health measurements. Consumer devices collect sensor data, which is then validated and labeled against gold-standard medical measurements obtained at medical facilities, creating a training dataset that enables consumer devices to achieve medical-grade accuracy through machine learning models.
Solution Approach 2:
The patent creates a copy of medical-grade measurement data by collecting and labeling consumer device sensor data against medical facility evaluation results. This labeled dataset serves as a surrogate training corpus that allows consumer devices to learn and replicate medical-grade measurement accuracy without requiring medical-grade hardware in every device.
2Productivity
If wearable devices collect more sensor data for training, then model performance is improved, but data labeling complexity and cost increase
Solution Approach 1:
The patent performs preliminary action by automatically detecting when users are at medical facilities using location data, and proactively collecting and labeling sensor data during these time periods. This automated preliminary labeling process eliminates the need for manual annotation of large datasets, reducing both complexity and cost while enabling accumulation of sufficient training data for high-performance models.
Solution Approach 2:
The system performs self-service by automatically identifying medical facility visits through GPS location data, retrieving medical evaluation results, and labeling corresponding sensor data without requiring manual intervention. This automated self-labeling process enables the system to accumulate large volumes of labeled training data efficiently, improving model performance without proportionally increasing labeling complexity.
3Measurement precision
If automatic labeling using medical facility data is implemented, then training data accuracy is improved, but system complexity increases
Solution Approach 1:
The patent implements a universal system where a single platform performs multiple functions: collecting sensor data from various consumer devices, detecting medical facility visits through location services, retrieving medical evaluation results, labeling data, and training machine learning models. This multi-functional system achieves high training data accuracy without requiring separate specialized systems for each function, thereby limiting the increase in overall system complexity.
Data Source
AI summary
Automated labeling of user sensor data is provided. It is determined that a user is at a medical facility using a location of a user device. User sensor data is collected from one or more user devices while the user is at the medical facility. A result is retrieved of a medical evaluation of the user performed at the medical facility. User sensor data collected during the medical evaluation is tagged with the retrieved result from the medical evaluation.