The project focuses on evaluating blood glucose level control methods in Type 1 diabetic patients. The research has explored the use of both online and offline reinforcement learning techniques to optimize glycemic control.
Phuwadol Viroonluecha's doctoral thesis at the Technical University of Cartagena explores the application of online and offline reinforcement learning for controlling blood glucose levels in Type 1 diabetic patients. The thesis provides insights into the effectiveness of these approaches in managing Type 1 diabetes.
A concept study undertaken by Guangyu Wang and colleagues aimed to optimize glycemic control in Type 2 diabetes using reinforcement learning, providing proof of concept for the applicability of these approaches in diabetes management.
This project focuses on the external administration of insulin necessary to maintain glucose levels in Type 1 diabetes. Researchers designed reinforcement learning algorithms to solve the control problem, emphasizing the key components of an Artificial Pancreas System (APS) for managing Type 1 diabetes.
A simulation of glycemic control in Type 1 diabetic patients is presented using a model inspired by existing works. The Python code illustrates a simulation based on the proposed model, and the results can be used to train reinforcement learning algorithms to optimize glycemic control in patients.
SimGlucose is a Python-based Type 1 diabetes simulator developed for research purposes. It is a Python version of the FDA-approved UVa/Padova simulator (2008 version) and includes 30 virtual patients, consisting of 10 adolescents, 10 adults, and 10 children.
The simulator follows the OpenAI Gym APIs for the simulation environment. It provides observations, rewards, termination signals, and information at each step, making it "reinforcement learning-ready".
Key features of SimGlucose include:
- Customizable reward function based on blood glucose measurements from the last hour.
- Incorporation of meals and physical activities into the model.
- Random and customizable scenario generator.
- Basic basal-bolus controller included, with a simple syntax for implementing custom controllers.
- Reproducible experiments through random seed specification.
- Performance analysis graphs, including blood glucose tracing.
To use SimGlucose, you can integrate it into your reinforcement learning pipeline to train models for optimizing glycemic control in Type 1 diabetic patients.
Following the glycemic control simulation in Type 1 diabetic patients, an environment based on the Actor Critic algorithm was built to train a model to optimize glycemic control by adjusting administered insulin levels.
The project represents a significant advancement in the autonomous management of diabetes, allowing for real-time, personalized adjustments to a patient's physiology with the goal of maintaining optimal glycemic control without constant human intervention.
+++