1 Temporal Abstractions for Interpreting Diabetic Patients Monitoring Data Advisor : Dr. Hsu Graduate : Min-Hong Lin IDSL seminar 2002/1/30.

Slides:



Advertisements
Similar presentations
Flexible and efficient retrieval of haemodialysis time series S. Montani, G. Leonardi, A. Bottrighi, L. Portinale, P. Terenziani DISIT, Sezione di Informatica,
Advertisements

Statistics Review – Part II Topics: – Hypothesis Testing – Paired Tests – Tests of variability 1.
ADVANCED STATISTICS FOR MEDICAL STUDIES Mwarumba Mwavita, Ph.D. School of Educational Studies Research Evaluation Measurement and Statistics (REMS) Oklahoma.
Monitoring diabetes Diabetes Outreach (March 2011)
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.
Outline input analysis input analyzer of ARENA parameter estimation
Experimental Design, Response Surface Analysis, and Optimization
Dynamic Bayesian Networks (DBNs)
Copyright (c) 2004 Brooks/Cole, a division of Thomson Learning, Inc. Chapter 13 Nonlinear and Multiple Regression.
Introduction of Probabilistic Reasoning and Bayesian Networks
Simulation Modeling and Analysis
Multi-Scale Analysis for Network Traffic Prediction and Anomaly Detection Ling Huang Joint work with Anthony Joseph and Nina Taft January, 2005.
Core Text Mining Operations 2007 년 02 월 06 일 부산대학교 인공지능연구실 한기덕 Text : The Text Mining Handbook pp.19~41.
Chapter 11 Multiple Regression.
Linear and generalised linear models
Chapter 2 Simple Comparative Experiments
The Analysis of Variance
Testing Bridge Lengths The Gadsden Group. Goals and Objectives Collect and express data in the form of tables and graphs Look for patterns to make predictions.
Chapter 9 Flashcards. measurement method that uses uniform procedures to collect, score, interpret, and report numerical results; usually has norms and.
Clinical Protocol Using Insulin Pump Easy Guideline for Initiating Insulin Pumps on Type 2 Diabetes Patients.
Data Mining : Introduction Chapter 1. 2 Index 1. What is Data Mining? 2. Data Mining Functionalities 1. Characterization and Discrimination 2. MIning.
2009 Mathematics Standards of Learning Training Institutes Algebra II Virginia Department of Education.
Aaker, Kumar, Day Ninth Edition Instructor’s Presentation Slides
Stat 1080 “Elementary Probability and Statistics” By Dr. AFRAH BOSSLY
Data Mining Chun-Hung Chou
Copyright © 2005 Brooks/Cole, a division of Thomson Learning, Inc Chapter 24 Statistical Inference: Conclusion.
Epidemiology The Basics Only… Adapted with permission from a class presentation developed by Dr. Charles Lynch – University of Iowa, Iowa City.
Presented by Dr. Soe Sandi Tint
Chapter 9 Hypothesis Testing and Estimation for Two Population Parameters.
Role of Statistics in Geography
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.
Lecture on Correlation and Regression Analyses. REVIEW - Variable A variable is a characteristic that changes or varies over time or different individuals.
Web Usage Mining for Semantic Web Personalization جینی شیره شعاعی زهرا.
Chapter 1 Introduction to Statistics. Statistical Methods Were developed to serve a purpose Were developed to serve a purpose The purpose for each statistical.
Educational Objectives
Data Preprocessing Dr. Bernard Chen Ph.D. University of Central Arkansas Fall 2010.
CEN st Lecture CEN 4021 Software Engineering II Instructor: Masoud Sadjadi Monitoring (POMA)
Enabling Reuse-Based Software Development of Large-Scale Systems IEEE Transactions on Software Engineering, Volume 31, Issue 6, June 2005 Richard W. Selby,
Detecting Group Differences: Mining Contrast Sets Author: Stephen D. Bay Advisor: Dr. Hsu Graduate: Yan-Cheng Lin.
Calculating Risk of Cost Using Monte Carlo Simulation with Fuzzy Parameters in Civil Engineering Michał Bętkowski Andrzej Pownuk Silesian University of.
Generic Tasks by Ihab M. Amer Graduate Student Computer Science Dept. AUC, Cairo, Egypt.
Intelligent Database Systems Lab 國立雲林科技大學 National Yunlin University of Science and Technology Mining Logs Files for Data-Driven System Management Advisor.
Chapter 3 System Performance and Models Introduction A system is the part of the real world under study. Composed of a set of entities interacting.
CHAPTER 2 Research Methods in Industrial/Organizational Psychology
STATISTICS AND OPTIMIZATION Dr. Asawer A. Alwasiti.
IMPORTANCE OF STATISTICS MR.CHITHRAVEL.V ASST.PROFESSOR ACN.
OPERATING SYSTEMS CS 3530 Summer 2014 Systems and Models Chapter 03.
Data Mining and Decision Support
1 SMU EMIS 7364 NTU TO-570-N Control Charts Basic Concepts and Mathematical Basis Updated: 3/2/04 Statistical Quality Control Dr. Jerrell T. Stracener,
1 Mining Episode Rules in STULONG dataset N. Méger 1, C. Leschi 1, N. Lucas 2 & C. Rigotti 1 1 INSA Lyon - LIRIS FRE CNRS Université d’Orsay – LRI.
Forecasting is the art and science of predicting future events.
The Need for Data Analysis 2 Managers track daily transactions to evaluate how the business is performing Strategies should be developed to meet organizational.
Simulation. Types of simulation Discrete-event simulation – Used for modeling of a system as it evolves over time by a representation in which the state.
CORRELATION-REGULATION ANALYSIS Томский политехнический университет.
Data Mining – Introduction (contd…) Compiled By: Umair Yaqub Lecturer Govt. Murray College Sialkot.
Chi Square Test Dr. Asif Rehman.
Modeling and Simulation CS 313
Lecture 1.31 Criteria for optimal reception of radio signals.
OPERATING SYSTEMS CS 3502 Fall 2017
INTRODUCTION AND DEFINITIONS
Chapter 7. Classification and Prediction
RAINGUAGE NETWORK DESIGN
Modeling and Simulation CS 313
A Growth Curve Analysis Participant Baseline Characteristics
CHAPTER 2 Research Methods in Industrial/Organizational Psychology
Chapter 12 Using Descriptive Analysis, Performing
Analysis of count data 1.
Measuring Data Quality and Compilation of Metadata
Discrete Event Simulation - 4
Process Capability.
Presentation transcript:

1 Temporal Abstractions for Interpreting Diabetic Patients Monitoring Data Advisor : Dr. Hsu Graduate : Min-Hong Lin IDSL seminar 2002/1/30

2 Outline Motivation Objective Background IDDM(Insulin Dependent Diabetes Mellitus) Temporal Abstractions TA for Interpreting Diabetic Patients Monitoring Data A Case Study Conclusions Comments

3 Motivation Several medical domains require: Analyzing and interpreting a large number of longitudinal data. Data coming from long-term monitoring of chronic patients. Physician must interpret information on the basis of a comprehensive analysis. In the interpretation task, the data pre-processing phase is a crucial step.

4 Objective To show that the data analysis and interpretation tasks can be carried out by transforming the raw data into a abstract view of the patient’s history. Propose a novel approach based on the combination of a temporal abstraction method with statistical and probabilistic techniques. Apply TA to the long-term monitoring of Insulin Dependent Diabetes Mellitus(IDDM) patients.

5 Insulin Dependent Diabetes Mellitus Diabetes Mellitus is a major chronic disease in developed countries (about 3%). Diabetes Mellitus is characterized by an alteration of the glucose metabolism due to a decreased endogenous production of insulin. In IDDM the patients must take exogenous insulin in order to prevent extremely high blood glucose levels(hyperglycemia).

6 Insulin Dependent Diabetes Mellitus(cont’d) The patients perform self-monitoring of the Blood Glucose Levels (BGL) and glycosuria at home, and report the monitoring and therapeutic data in a diary. The accuracy of the patients’ self-care is very important, since the onset and development of diabetic complications is strictly related to the degree of metabolic control. The physicians revise the therapy during periodical visits, every two/six months.

7 IT for Diabetes Care To improve the quality of the therapy revision process and the patients’ management, several computer-based systems have been proposed since the early 80’s. Different systems categories: Day-by-day advisory systems V.S. visit-by visit advisory systems Model-based systems V.S. Data-driven systems Statistical analysis and graphical representations Time-series analysis and temporal abstractions

8 Temporal Abstractions TA is an AI methodology. TAs are used in data interpretation to solve the temporal abstraction task whose goal is to abstract high level concepts from time-stamped data. In the medical domain, TAs can be used to describe patients states holding over time periods. The principle of the TA method is to move from a time-point to an completely interval-based representation of the monitoring data.

9 Temporal Abstractions(cont’d) TA task is decomposed into two main type of TA subtasks: Basic TA: solved by mechanisms that abstract time- stamped data into intervals(input data are events and outputs are episodes) State TAs: detect episodes associated to qualitative levels of time-varying variables, like normal or abnormal states. Trend TAs: detect patterns like increase, decrease and stationarity in a numerical time series. Complex TA: solved by mechanism that abstract intervals into other intervals(input and output data are episodes)

10 TA method ontology

11 Basic TA task For each basic TA it defines two parameters: Granularity: represents the maximum allowed temporal gap between two measurements that can be aggregated into the same episode. Minimum extent: represents the minimum time span of an episode to be considered relevant.

12 An example of the basic TA task

13 Complex TA task The mechanism solving the complex TA task searches for temporal relationships between episodes. The temporal relationships investigated can be expressed through temporal operators defined in the Allen algebra(before, after, meets, overlaps, starts, finishes, equals, during)

14 An example of the complex TA task

15 An example of the complex TA task(cont’d) If the input episodes refer to patterns extracted from different time series, the method can detect patterns in multi-dimensional data. For example, the problem of investigating whether a persistent cough and high fever occur simultaneously in a patient’s history. PERSISTENT COUGH OVERLAPS HIGH FEVER

16 Temporal Abstraction for the Interpretation of Diabetic Patients’ Monitoring Data Data Pre-Processing Analyze the original time-series using TAs Derive a collection of new time-series by computing the TAs that are true in each time point of the original time scale. Data Interpretation

17 Data Pre -Processing Subdivide the 24-hours daily period into a set of consecutive non-overlapping time slices Perform the analysis on the time series of three variables ( BGL, glycosuria and insulin dosages)

18 Data Pre -Processing(cont’d) Define a set of basic and complex abstractions for each time slice.

19 Data Pre -Processing(cont’d) Characterize the patients’ behavior through the concept of Abstract State(ABST), that corresponds to the combination of the TAs that are true in that period. The general form of the abstract state in the i-th day for the j-th time-slice(ABST ij ) is:

20 An example of data pre-processing

21 An example of data pre-processing(cont’d)

22 An example of data pre-processing(cont’d)

23 Data Interpretation The Blood Glucose Modal Day(BG-MD) The Blood Glucose Modal Daily Pattern(BG-MDP) Exploiting TA Time-Spans Exploiting Complex TAs

24 The Blood Glucose Modal Day(BG-MD) The BG-MD is a characteristic daily BGL pattern that summarizes the typical patient’s response to the therapy in a specific monitoring period It is usually derived as the collection of the most probable blood glucose qualitative levels in each time slice. K:states, N: monitoring days, D:collect measurements, M=N-D: missing data, d l is the number of occurrence of the l-th state in the monitoring period. Ignorance (IG) in the monitoring period: The modal day can be extracted by taking the BGL states with the highest p inf in each time slice.

25 The Blood Glucose Modal Daily Pattern(BG-MDP) BG-MDP is the most frequent sequence of abstract states of the BGL variable in the different time slices of one day. Search parameters: the maximum allowed ignorance (MIG) in one time slice (I.e. the maximum number of allowed missing data) and the minimum probability bound for the joint probability distribution P inf. Time slices selection: only the time slices that have ignorance level lower than MIG will include in the BG- MDP Search: perform an exhaustive search of the daily patterns that have a lower probability bound higher than P inf.

26 Exploiting TA Time-Spans If the time span of the BGL normoglycemic episodes have an exponential distribution, it is clear that the patient is not able to control his/her glucose metabolism for a long period. Perform a non-linear least-squares fitting of the model Once we have estimated λ, we test the hypothesis that the data follows an exponential law with parameter through the x 2 statistics. The exponential distribution hypothesis is rejected with degree α if x 2 h-1 > x 2 h-1 (1- α)

27 Exploiting Complex TAs In clinical practice, the physician usually tries to combine the information coming from the different variables under monitoring. Somogyi effect : is detected by looking for “hyperglycemia at Breakfast with absence of glycosuria” Dawn effect : is detected by searching for “hyperglycemia at Breakfast with presence of glycosuria” Metabolic Instability : in which a “BGL increase” is immediately followed by a “BGL decrease ” or vice- versa.

28 An example of Exploiting Complex TAs

29 A Case Study A 14 years-old female patient, monitored for a period of 165 days. The BGL data are show below:

30 A Case Study (cont’d)

31 A Case Study (cont’d)

32 A Case Study (cont’d)

33 A Case Study (cont’d)

34 A Case Study (cont’d)

35 A Case Study (cont’d)

36 A Case Study (cont’d)

37 Conclusions TA method requires an intensive knowledge acquisition effort for the TA definition, both in terms of types of TAs to be applied to the particular problem and in terms of the parameters that some TAs require. TA method will enable the assessment of a new class of data mining systems.

38 Comments How to prevent from diabetic: Diet Physical exercise Check the BGL regularly