Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Attention Networks for Personalized Mealtime Insulin Dosing in People with Type 1 Diabetes

Anas El Fathi, Elliott Pryor, Marc D. Breton

TL;DR

This work demonstrates how a reinforcement learning agent, employing a self-attention encoder network, can effectively mimic and enhance this intuitive process, offering the potential to simplify treatment and promising improved quality of life and glycemic outcomes for people with T1D.

Abstract

Calculating mealtime insulin doses poses a significant challenge for individuals with Type 1 Diabetes (T1D). Doses should perfectly compensate for expected post-meal glucose excursions, requiring a profound understanding of the individual's insulin sensitivity and the meal macronutrients'. Usually, people rely on intuition and experience to develop this understanding. In this work, we demonstrate how a reinforcement learning agent, employing a self-attention encoder network, can effectively mimic and enhance this intuitive process. Trained on 80 virtual subjects from the FDA-approved UVA/Padova T1D adult cohort and tested on twenty, self-attention demonstrates superior performance compared to other network architectures. Results reveal a significant reduction in glycemic risk, from 16.5 to 9.6 in scenarios using sensor-augmented pump and from 9.1 to 6.7 in scenarios using automated insulin delivery. This new paradigm bypasses conventional therapy parameters, offering the potential to simplify treatment and promising improved quality of life and glycemic outcomes for people with T1D.

Attention Networks for Personalized Mealtime Insulin Dosing in People with Type 1 Diabetes

TL;DR

This work demonstrates how a reinforcement learning agent, employing a self-attention encoder network, can effectively mimic and enhance this intuitive process, offering the potential to simplify treatment and promising improved quality of life and glycemic outcomes for people with T1D.

Abstract

Calculating mealtime insulin doses poses a significant challenge for individuals with Type 1 Diabetes (T1D). Doses should perfectly compensate for expected post-meal glucose excursions, requiring a profound understanding of the individual's insulin sensitivity and the meal macronutrients'. Usually, people rely on intuition and experience to develop this understanding. In this work, we demonstrate how a reinforcement learning agent, employing a self-attention encoder network, can effectively mimic and enhance this intuitive process. Trained on 80 virtual subjects from the FDA-approved UVA/Padova T1D adult cohort and tested on twenty, self-attention demonstrates superior performance compared to other network architectures. Results reveal a significant reduction in glycemic risk, from 16.5 to 9.6 in scenarios using sensor-augmented pump and from 9.1 to 6.7 in scenarios using automated insulin delivery. This new paradigm bypasses conventional therapy parameters, offering the potential to simplify treatment and promising improved quality of life and glycemic outcomes for people with T1D.
Paper Structure (18 sections, 8 equations, 4 figures, 2 tables)

This paper contains 18 sections, 8 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Background and Motivation
  3. Traditional Bolus Calculators for T1D
  4. Review of RL algorithms for Bolus Calculators
  5. Review of Proximal Policy Optimization
  6. Methods
  7. Problem formulation
  8. Neural Network Architecture
  9. Observations Preprocessing
  10. Encoded States
  11. Actions
  12. Rewards
  13. Experimental Setup and Training
  14. Outcomes Metrics
  15. Results
  16. ...and 3 more sections

Figures (4)

  • Figure 1: Steps of processing the observations: (i) data is normalized and decayed to emphasize the slow dynamics of insulin and meal, (ii) data is sliced using a window $w$ and stride $s$, (iii) data is rearranged to align periodic events.
  • Figure 2: Neural networks architecture including (1) the encoder network (specifically the self-attention network) (2) the value network (3) the actor network.
  • Figure 3: Depiction of reward calculation in equation \ref{['eqn:reward']}. The smaller the shaded grey areas (1) and (2) the bigger the reward.
  • Figure 4: Test experiment of 20 VS. The same meal (60g) is given 3-times a day. Insulin sensitivity is doubled on day 3 and halved on day 7.