Reinforcement Learning Insulin Dosing Recommendation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current closed-loop systems for managing blood glucose in diabetes patients are inaccurate due to insufficient consideration of all influencing parameters, leading to unpredictable insulin dosing and potential hypoglycemia or hyperglycemia.
Innovation Solution
A control device using a reinforcement learning algorithm that retrieves user data, including drug infusion amounts, physiological values, and estimated values, to determine a recommendation value for insulin dosing, with a configuration that adapts and improves accuracy over time, incorporating a reward score system to refine initial parameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If time-based data is used to predict future glycemia, then the system can operate with simple data collection, but the prediction accuracy is insufficient leading to irrelevant insulin dosing
Solution Approach 1:
The patent changes the parameters used for prediction from simple time-based data to multiple physiological parameters including glycemia, insulin dosage, carb intake, physical activity, stress levels, and sleep patterns. This parameter transformation enables the system to capture comprehensive influences on glycemia while maintaining operational simplicity through automated multi-parameter integration.
Solution Approach 2:
The system integrates multiple data sources and calculation methods into a single unified prediction model that handles various physiological influences simultaneously. The controller performs multiple functions including predicting future glycemia, calculating carb intake, assessing physical activity impact, and determining stress effects all within one comprehensive framework.
2Measurement precision
If all influencing parameters are considered, then glycemia prediction accuracy improves, but the system complexity increases
Solution Approach 1:
The patent segments the complex prediction system into distinct functional modules: glycemia prediction module, carb intake calculation module, physical activity assessment module, stress evaluation module, and insulin dosage calculation module. Each module processes specific parameters independently before integrating results, making the overall system more manageable while maintaining comprehensive parameter consideration.
Solution Approach 2:
The system automatically collects, processes, and integrates all relevant physiological parameters without requiring manual input or external intervention. The controller self-adjusts by continuously learning from user feedback and automatically updating prediction models, eliminating the need for complex manual configuration while maintaining high accuracy.
3Measurement precision
If reinforcement learning algorithm is used to iteratively improve recommendations, then dosing accuracy improves over time, but computational complexity increases
Solution Approach 1:
The system implements a feedback loop where the controller continuously monitors actual glycemia results, compares them with predicted values, and uses this information to adjust the reinforcement learning algorithm. User feedback on dosing effectiveness automatically refines the model parameters, enabling iterative improvement of accuracy while keeping computations manageable through targeted parameter adjustment.
Solution Approach 2:
The reinforcement learning algorithm dynamically adjusts computational parameters based on available data and prediction confidence levels. The system changes calculation complexity adaptively, using simpler models when data is limited and more complex models as more historical data becomes available, thereby managing computational complexity while improving dosing accuracy over time.
Data Source
Figure 1~2
AI summary
A control device (30) for determining a recommendation value of a control parameter of a fluid infusion device (20). The control device (30) comprises a retrieving unit (32) configured to retrieve user data. Each data of the user data having a timestamp and the user data being related to a unique user. The user data comprises at least a plurality of amounts of a drug infused to the unique user; a plurality of physiological values of the unique user; and a plurality of estimated values. The control device (30) also comprises a recommendation unit (34). The recommendation unit (34) is configured to determine the recommendation value based at least on a data of the user data and using a reinforcement learning algorithm comprising a plurality of initial reinforcement parameters.