Meal Detection and Meal Size Estimation for Type 1 Diabetes: a Variable State Dimension Approach 2015 DSCC Columbus Oct. 30 Jinyu Xie, Prof. Qian Wang.

Slides:



Advertisements
Similar presentations
TARGET DETECTION AND TRACKING IN A WIRELESS SENSOR NETWORK Clement Kam, William Hodgkiss, Dept. of Electrical and Computer Engineering, University of California,
Advertisements

Probabilistic Reasoning over Time
A Sensor Fault Diagnosis Scheme for a DC/DC Converter used in Hybrid Electric Vehicles Hiba Al-SHEIKH Ghaleb HOBLOS Nazih MOUBAYED.
Data Mining Methodology 1. Why have a Methodology  Don’t want to learn things that aren’t true May not represent any underlying reality ○ Spurious correlation.
U D A Neural Network Approach for Diagnosis in a Continuous Pulp Digester Pascal Dufour, Sharad Bhartiya, Prasad S. Dhurjati, Francis J. Doyle III Department.
Fault-Tolerant Target Detection in Sensor Networks Min Ding +, Dechang Chen *, Andrew Thaeler +, and Xiuzhen Cheng + + Department of Computer Science,
Blood Glucose Prediction using Physiological Models and Support Vector Regression Razvan Bunescu Nigel Struble Cindy Marling Ohio University, Athens, OH.
STAT 497 APPLIED TIME SERIES ANALYSIS
Mixed Model Analysis of Highly Correlated Data: Tales from the Dark Side of Forestry Christina Staudhammer, PhD candidate Valerie LeMay, PhD Thomas Maness,
Classical inference and design efficiency Zurich SPM Course 2014
Sam Pfister, Stergios Roumeliotis, Joel Burdick
280 SYSTEM IDENTIFICATION The System Identification Problem is to estimate a model of a system based on input-output data. Basic Configuration continuous.
Cost-Sensitive Classifier Evaluation Robert Holte Computing Science Dept. University of Alberta Co-author Chris Drummond IIT, National Research Council,
Engineering Data Analysis & Modeling Practical Solutions to Practical Problems Dr. James McNames Biomedical Signal Processing Laboratory Electrical & Computer.
Darlene Goldstein 29 January 2003 Receiver Operating Characteristic Methodology.
1 Adaptive Kalman Filter Based Freeway Travel time Estimation Lianyu Chu CCIT, University of California Berkeley Jun-Seok Oh Western Michigan University.
Student: Hsu-Yung Cheng Advisor: Jenq-Neng Hwang, Professor
Energy-efficient Self-adapting Online Linear Forecasting for Wireless Sensor Network Applications Jai-Jin Lim and Kang G. Shin Real-Time Computing Laboratory,
1 Progress Towards an Artificial Pancreas for T1D WILLIAM TAMBORLANE, MD Chief of Pediatric Endocrinology, Yale University, Deputy Director, Yale Center.
Suhyla Alam (Eastern Virginia Medical School), Amy West, Maura Downey, Jane EB Reusch, Kristen Nadeau University of Colorado Denver and Children’s Hospital.
Multiantenna-Assisted Spectrum Sensing for Cognitive Radio
Industrial Process Modelling and Control Ton Backx Emeritaatsviering Joos Vandewalle.
Real-time control Model-Based Blood Glucose Control for Neonatal Intensive Care Engineering robust solutions for our most fragile infants Background and.
Diabetes Technology Update
Mealtime Glycemic Excursions in Pediatric Subjects with Type 1 Diabetes: Results of the Diabetes Research in Children (DirecNet) Accuracy Study Study Group.
1 Clinical Investigation and Outcomes Research Statistical Issues in Designing Clinical Research Marcia A. Testa, MPH, PhD Department of Biostatistics.
Background: DirecNet Diabetes Research in Children Network NIH funded collaborative study group 5 clinical centers, central laboratory, coordinating center,
Soft Sensor for Faulty Measurements Detection and Reconstruction in Urban Traffic Department of Adaptive systems, Institute of Information Theory and Automation,
Safety of Outpatient Closed-Loop Control: First Randomized Crossover Trials of a Wearable Artificial Pancreas Featured Article: Boris P. Kovatchev, Eric.
Introduction Clinical Setting - Glucoregulatory system - Patients and Data Assessment Proc. - Glucose sensors - Glycemia control system Control System.
Artificial Pancreas Project at Cambridge R. Hovorka, J.M. Allen, L.J.Chassin, A. De Palma, D. Elleri, J. Harris, J.F. Hayes, T. Hovorka, K. Kumareswaran,
BY IRFAN AZHAR Control systems. What Do Mechatronics Engineers Do?
Time Series Data Analysis - I Yaji Sripada. Dept. of Computing Science, University of Aberdeen2 In this lecture you learn What are Time Series? How to.
Chapter 21 R(x) Algorithm a) Anomaly Detection b) Matched Filter.
Prognosis of Gear Health Using Gaussian Process Model Department of Adaptive systems, Institute of Information Theory and Automation, May 2011, Prague.
Karman filter and attitude estimation Lin Zhong ELEC424, Fall 2010.
Continuous Glucose Monitors
Lecture 8 Simple Linear Regression (cont.). Section Objectives: Statistical model for linear regression Data for simple linear regression Estimation.
BAGGING ALGORITHM, ONLINE BOOSTING AND VISION Se – Hoon Park.
L Berkley Davis Copyright 2009 MER301: Engineering Reliability Lecture 9 1 MER301:Engineering Reliability LECTURE 9: Chapter 4: Decision Making for a Single.
Long Term Verification of Glucose-Insulin Regulatory System Model Dynamics THE 26th ANNUAL INTERNATIONAL CONFERENCE OF THE IEEE ENGINEERING IN MEDICINE.
Effectiveness of Early Intensive Therapy On β-Cell Preservation in Type 1 Diabetes Featured Article: Bruce Buckingham, M.D., Roy W. Beck, M.D., P.H.D.,
© Copyright McGraw-Hill Correlation and Regression CHAPTER 10.
Stable Multi-Target Tracking in Real-Time Surveillance Video
ABSTRACT Hyperglycaemia is prevalent in critical care, and tight control reduces mortality. Targeted glycaemic control can be achieved by frequent fitting.
A Trust Based Distributed Kalman Filtering Approach for Mode Estimation in Power Systems Tao Jiang, Ion Matei and John S. Baras Institute for Systems Research.
WCRP Extremes Workshop Sept 2010 Detecting human influence on extreme daily temperature at regional scales Photo: F. Zwiers (Long-tailed Jaeger)
Xiali Hei, Xiaojiang Du, Shan Lin Temple University
Brain-Machine Interface (BMI) System Identification Siddharth Dangi and Suraj Gowda BMIs decode neural activity into control signals for prosthetic limbs.
September 28, 2000 Improved Simultaneous Data Reconciliation, Bias Detection and Identification Using Mixed Integer Optimization Methods Presented by:
1 Use of Multiple Integration and Laguerre Models for System Identification: Methods Concerning Practical Operating Conditions Yu-Chang Huang ( 黃宇璋 ) Department.
Chapter 13 Understanding research results: statistical inference.
The Unscented Kalman Filter for Nonlinear Estimation Young Ki Baik.
Robust Localization Kalman Filter & LADAR Scans
MPC for Type 1 Diabetes Based on Personalized LTV Model Consisting of both Insulin and Meal inputs Qian Wang, Jinyu Xie et al Background Modeling Control.
State-Space Recursive Least Squares with Adaptive Memory College of Electrical & Mechanical Engineering National University of Sciences & Technology (NUST)
TEMPLATE DESIGN © CONTINUOUS GLUCOSE MONITORING, ORAL GLUCOSE TOLERANCE, AND INSULIN – GLUCOSE PARAMETERS IN ADOLESCENTS.
Lauren M. Huyett, Eyal Dassau, and Francis J. Doyle III Department of Chemical Engineering University of California Santa Barbara Santa Barbara, CA
A Telemedicine System for Modeling and Managing Blood Glucose David L. Duke October 26, 2009 Intelligent Diabetes Assistant.
ASEN 5070: Statistical Orbit Determination I Fall 2014
M. Kuhn, P. Hopchev, M. Ferro-Luzzi
Copyright © 2004 American Medical Association. All rights reserved.
NOT A NEW CONCEPT: Really ‘Back to the Future’
Utilizing New Technologies and Insulins for a Tighter Grip on PPG
Cox Regression Model Under Dependent Truncation
Principles of the Global Positioning System Lecture 11
Blood/Subcutaneous Glucose Dynamics Estimation Techniques
Kalman Filter: Bayes Interpretation
Glucose control performance (by CGM) characterized by median and interquartile range cumulative % time in glucose range (A), overall glucose (B), and insulin.
(A) Mean glucose concentrations (standard error) over a 3-hour period in 21 placebo- and 15 pramlintide-treated patients with type 1 diabetes treated for.
Presentation transcript:

Meal Detection and Meal Size Estimation for Type 1 Diabetes: a Variable State Dimension Approach 2015 DSCC Columbus Oct. 30 Jinyu Xie, Prof. Qian Wang Department of Mechanical and Nuclear Engineering The Pennsylvania State University, University Park

Outlines Motivation Modeling Variable State Dimension Approach Meal Detection and Meal Size Estimation Simulation Results Conclusion 2 of 24

Motivation Controllers with meal or manual meal bolus showed larger amount of time in target range (70 – 180 mg/dL) than fully automated controllers [1] [2]. A study by Burdick et al[3]: a group of young patients aged from 7 to 20 missed 2 mealtime boluses per week on average. Artificial Pancreas Patient Glucose Sensor Model and Controller Bio-Mechanical Interface Insulin Pump Meal Announcement (Feed-forward) Insulin Delivery Insulin Dosage Glucose Measurement Data Processing Insulin Bolus = Meal/IC Ratio Meal Detection & Estimation 3 of 24

Literature Review 4 of 24

General idea Insulin Meal Internal Factors (Renal, Liver, etc.) Blood Glucose (BG) Measurement = Insulin Effect Meal Intake Effect + + Noise + Same Insulin Injection History A general glucose dynamic model: For models without meal input, meal detection is possible while meal size estimation is not available. The accuracy of meal size estimation depends on the quality of the model. 5 of 24

- Glucose Autoregressive Component - Insulin Regression Component - Carbohydrate Regression Component Modeling Model Parameters: ζ : True Blood Glucose (BG) y : BG Measurement v : Filtered Insulin Input w: Filtered Carbohydrate Input η : Process Noise (model noise) ξ : Measurement Noise 6 of 24 A linear regressive model

Modeling FIR Filter for Insulin/Carbohydrate Absorption Profile Unit Impulse Response General Shaping. Each patient can be personalized by 7 of 24

Modeling Parameter Training Scenario, suggested by Parcival et al [9] Data generated by the FDA approved UVa/Padova Simulator 8 of 24

StateInsulin EffectMeal Effect State Space Representation are assumed to be White Gaussian Noise. ζ : True Blood Glucose (BG) y : BG Measurement v : Filtered Insulin Input w: Filtered Carbohydrate Input η : Process Noise ξ : Measurement Noise Parameters obtained by the training. 9 of 24

Monitoring the innovation (residual) Variable State Dimension (VSD) Quiescent Model Maneuvering Model Maneuver Detected Maneuver Terminated - An Adaptive Estimation Algorithm Linear Kalman Filter is run and switched between the two models. Monitoring the augmented disturbance 10 of 24

Algorithm Overview Monitoring a sliding window of data. Quiescent Model. Monitoring the innovation Maneuvering Model. Monitoring the augmented disturbance Meal occurrence time is assumed to be at the beginning of the sliding window. Meal Size Estimation starts right after the meal is detected, and evolves as new observation comes. 11 of 24

Meal (Maneuver) Detection Monitoring the innovation Hypothesis Test : is White Gaussian Process. Rectangular Weighted Window Exponential Weighted Window Test Statistics with sliding window Window Length = s - Fading Factor Effective Window Length where 12 of 24 where

Meal (Maneuver) Detection (cont’d) is rejected, if, where c is the confidence level. Hypothesis Test : is White Gaussian Process. If is true, Quiescent Maneuvering H 0 rejected H 0 accepted 13 of 24 Meal is detected when H 0 is rejected

Kalman Filter Initialization at Maneuvering Stage Maneuver Detected Assumed maneuver onset time - Estimation with quiescent model - Estimation with maneuvering model of 24 Estimated State and State Covariance Initialization:

Kalman Filter Initialization at Maneuvering Stage (cont’d) 15 of 24 State Variance Initialization:

Meal Size Estimation Assumptions: Only one meal for one maneuver (multiple meals will be considered as one meal) The meal is an impulse lasting only one time step Meal Effect Γ(w(t)) Disturbance d(t) = 16 of 24

Maneuver Termination Monitoring the augmented disturbance state Hypothesis Test H 0 : is a White Gaussian Process If is true, H 0 is accepted, if, where c is the confidence level. where Effective Window Length Maneuvering Quiescent H 0 rejected H 0 accepted 17 of 24 Maneuver is terminated when H 0 is accepted

Simulation Platform Data generated by the FDA approved UVa/Padova simulator: Noise-free Plasma Blood Glucose (Plasma) Glucose measured by Continuous Glucose Monitor (CGM) 18 of patients, 48 hour

Simulation Results - Plasma Adolescent #005 - Plasma Equal trust on the model and the measurement. Without meal information With meal information 19 of 24 VSDIGIMG 0.52 (0.34)0.28 (0.52)0.45 (0.43) All 30 subjects Mean (Standard Deviation) Red - No meal information, Black – full meal information

Simulation Results - CGM Adolescent #005 - CGM Relying more on the model (due to the sensor noise) Without meal information With meal information 20 of 24 VSDIGIMG 0.35 (0.34)-0.07 (0.77)0.37 (0.44) All 30 subjects Mean (Standard Deviation) Red - No meal information, Black – full meal information

Meal Detection Performance Dassau et al (CGM) [4] Lee & Bequette (Not known) [6] Cameron & Niemeyer (CGM) [8] VSD (Plasma) VSD (CGM) Meal detection time since a meal starts (min) (±10)34(±19) Meal detection time since meal onset* (min) -3181(±10)8(±19) Sensitivity -82%-97%95% False alarm rate -7.6%-10%17% * Meal onset is defined as the time when the glucose level rises by 4mg/dL after meal [6]. For our data, average meal onset time: 21 min (Plasma), 26 min (CGM) Sensitivity = TP/(TP+FN) Patient had a mealPatient did not have a meal Detector says yesTrue Positive (TP)False Positive (FP) Detector says noFalse Negative (FN)True Negative (TN) False alarm rate = FP/(TP+FP) 21 of 24

Meal Size Estimation Performance 22 of 24

Conclusion and Future Work A Kalman Filter switches between quiescent state model and maneuvering state model (an adaptive estimator) 86% increase in goodness of fit (R 2 ) compared to Kalman Filter without model switching (both no meal information available) Better or comparable detection performance <13% meal size estimation error rate Future work on online estimation and control Possible extension to detect and estimate physical activities 23 of 24

Reference [1] Doyle, Francis J., Lauren M. Huyett, Joon Bok Lee, Howard C. Zisser, and Eyal Dassau. "Closed-loop artificial pancreas systems: engineering the algorithms." Diabetes care 37, no. 5 (2014): [2] Wang, Q., J. Xie, P. Molenaar, and J. Ulbrecht. “Model predictive control for type 1 diabetes based on personalized linear time-varying subject model consisting of both insulin and meal inputs: an in silico evaluation.” Journal of Diabetes Science and Technology, 9(4) (2015): [3] Burdick, Jonathan, H. Peter Chase, Robert H. Slover, Kerry Knievel, Laura Scrimgeour, Aristides K. Maniatis, and Georgeanna J. Klingensmith. "Missed insulin meal boluses and elevated hemoglobin A1c levels in children receiving insulin pump therapy." Pediatrics 113, no. 3 (2004): e221-e224. [4] Dassau, Eyal, B. Wayne Bequette, Bruce A. Buckingham, and Francis J. Doyle. "Detection of a Meal Using Continuous Glucose Monitoring Implications for an artificial β-cell." Diabetes care 31, no. 2 (2008): [5] Lee, Hyunjin, and B. Wayne Bequette. "A closed-loop artificial pancreas based on model predictive control: Human-friendly identification and automatic meal disturbance rejection." Biomedical Signal Processing and Control 4.4 (2009): [6] Lee, Hyunjin, Bruce A. Buckingham, Darrell M. Wilson, and B. Wayne Bequette. "A closed-loop artificial pancreas using model predictive control and a sliding meal size estimator." Journal of diabetes science and technology 3, no. 5 (2009): [7] Cameron, Fraser, Günter Niemeyer, and Bruce A. Buckingham. "Probabilistic evolving meal detection and estimation of meal total glucose appearance."Journal of diabetes science and technology 3.5 (2009): [8] Fraser Cameron and Gunter Niemeyer. Predicting blood glucose levels around meals for patients with type i diabetes. In ASME 2010 Dynamic Systems and Control Conference, pages 289–296. American Society of Mechanical Engineers, [9] Matthew W Percival, Wendy C Bevier, Youqing Wang, Eyal Dassau, Howard C Zisser, Lois Jovanoviˇc, and Francis J Doyle. Modeling the effects of subcutaneous insulin administration and carbohydrate consumption on blood glucose. Journal of diabetes science and technology, 4(5):1214–1228, of 24

Online Basal-Bolus Control 25 of 24