Related Experiment Video
Updated: Jun 22, 2025
12:08
Improving IV Insulin Administration in a Community Hospital
Published on: June 11, 2012
18.8K
An automatic deep reinforcement learning bolus calculator for automated insulin delivery systems
Sayyar Ahmad1, Aleix Beneyto1, Taiyu Zhu2
1Modeling and Intelligent Control Engineering Laboratory, Institute of Informatics and Applications, University of Girona, 17003, Girona, Spain.
Scientific Reports
|July 2, 2024
Summary
This study introduces a fully automatic insulin delivery (FAID) system using deep reinforcement learning to manage type 1 diabetes without requiring meal announcements, improving glycemic control.
Area of Science:
- Biomedical Engineering
- Artificial Intelligence in Medicine
- Endocrinology
Background:
- Hybrid automatic insulin delivery (HAID) systems require patient meal announcements for effective glycemic control in type 1 diabetes.
- Carbohydrate estimation in HAID systems is error-prone and burdensome for patients.
- Current systems struggle with unannounced meals, impacting glucose management.
Purpose of the Study:
- To develop and evaluate a fully automatic insulin delivery (FAID) system that eliminates the need for patient meal announcements.
- To utilize deep reinforcement learning (DRL) for calculating insulin bolus without carbohydrate information.
- To compensate for unannounced meals and improve glycemic control in type 1 diabetes.
Main Methods:
- A deep reinforcement learning (DRL) algorithm was developed to calculate insulin bolus for unannounced meals.
- The DRL bolus calculator was integrated with a closed-loop controller and a meal detector.
- In-silico trials were conducted using 68 virtual patients based on the modified UVa/Padova simulator.
Main Results:
- The FAID system demonstrated improved time in the target glucose range (70-180 mg/dL) compared to a standard bolus calculator (SBC) in HAID systems, even with carbohydrate misestimation.
- FAID showed comparable or superior performance to HAID systems in maintaining glycemic control.
- The DRL-based approach effectively compensated for unannounced meals without requiring carbohydrate intake information.
Conclusions:
- The proposed fully automatic insulin delivery (FAID) system effectively manages glycemic control in type 1 diabetes without patient intervention for meal announcements.
- Deep reinforcement learning offers a promising approach for autonomous insulin delivery, overcoming limitations of current HAID systems.
- FAID systems have the potential to significantly improve the quality of life for individuals with type 1 diabetes.
Related Concept Videos
01:16Insulin: Dosing Regimen and Adverse Effects
165
Insulin-replacement therapy usually includes both long-acting insulin (basal) and short-acting insulin (to cater to postprandial needs). In a diverse group of type 1 diabetes patients, the average daily insulin dose is typically 0.5-0.7 units/kg body weight. However, obese patients and pubertal adolescents may need more due to insulin resistance.
The basal dose constitutes about 40%-50% of the total daily dose, with the rest as premeal insulin. The mealtime insulin dose should mirror...
The basal dose constitutes about 40%-50% of the total daily dose, with the rest as premeal insulin. The mealtime insulin dose should mirror...
165
01:27Insulin Formulations: Types and Delivery
184
Insulin preparations are categorized by their duration of action into short-acting and long-acting types. Two strategies are used to modify insulin's absorption and pharmacokinetic profile: slowing the absorption post-subcutaneous injection, or altering human insulin's amino acid sequence or protein structure. These changes retain the insulin's ability to bind to the insulin receptor, but alter its behavior in solution or after injection.
Short-acting insulins are divided into...
Short-acting insulins are divided into...
184
01:19One-Compartment Open Model for IV Bolus Administration: General Considerations
184
The one-compartment model is a pharmacokinetic tool that models the body as a single, uniform compartment, facilitating the understanding of drug distribution and elimination. This model is particularly beneficial for intravenous (IV) bolus administration, where the drug rapidly circulates throughout the body.
The drug's presence in the body is defined by an equation representing the difference between the rates of drug entry and exit. Key parameters—elimination rate constant,...
The drug's presence in the body is defined by an equation representing the difference between the rates of drug entry and exit. Key parameters—elimination rate constant,...
184
01:09One-Compartment Open Model for IV Bolus Administration: Estimation of Elimination Rate Constant, Half-Life and Volume of Distribution
237
The one-compartment open model is a simplified approach used in pharmacokinetics to understand the distribution and elimination of a drug administered through an intravenous bolus. This model assumes rapid drug dispersal throughout the body and elimination using a first-order process. Key pharmacokinetic parameters, such as the elimination rate constant (k), half-life (t1/2), and the apparent volume of distribution (Vd), can be estimated from this model. The elimination rate is calculated...
237
01:18Two-Compartment Open Model: IV Bolus Administration
463
The two-compartment model for intravenous (IV) bolus administration illustrates drug distribution in the body, subdividing it into central and peripheral compartments. This model operates on the concept of two-compartment kinetics. The drug's plasma concentration shows a bi-exponential decline following IV bolus administration, signaling the presence of two disposition processes: distribution and elimination.
The disparity between drug input and the sum of drug transfer rates between...
The disparity between drug input and the sum of drug transfer rates between...
463
00:56One-Compartment Open Model for IV Bolus Administration: Estimation of Clearance
72
Clearance is a key pharmacokinetic parameter that quantifies the volume of body fluid from which a drug is entirely removed within a specific time frame. It is crucial in assessing how a drug is eliminated from the body and has critical clinical applications.
In the one-compartment open model for intravenous (IV) bolus administration, clearance is estimated by dividing the elimination rate by the plasma drug concentration. This equation leverages the elimination rate constant and the apparent...
In the one-compartment open model for intravenous (IV) bolus administration, clearance is estimated by dividing the elimination rate by the plasma drug concentration. This equation leverages the elimination rate constant and the apparent...
72