Automatic Detection of Excessive - PowerPoint Presentation

Slide1

Automatic Detection of Excessive Glycemic Variability for Diabetes Management

Matthew Wiley, Razvan Bunescu, Cindy Marling,Jay Shubrook and Frank SchwartzSchool of Electrical Engineering and Computer ScienceAppalachian Rural Health Institute Diabetes and Endocrine CenterAthens, Ohio, USA

Automatic Detection of Excessive

Glycemic

Variability for Diabetes Management Wiley et al.

1Slide2

DiabetesBody fails to effectively produce and/or use insulin

Treated and managed with blood glucose controlAutomatic Detection of Excessive Glycemic Variability for Diabetes Management Wiley et al.2Slide3

DiabetesBody fails to effectively produce and/or use insulin

Treated and managed with blood glucose control346 million people world wideTwo major types:Type IType IIAutomatic Detection of Excessive Glycemic Variability for Diabetes Management Wiley et al.

3Slide4

Poor Control Increases Risk of Complications

Automatic Detection of Excessive Glycemic Variability for Diabetes Management Wiley et al.4

CONTROL

Foot Ulcers

Angina Heart Attack

Coronary Bypass

Surgery Stroke

Blindness

Amputation Dialysis

Kidney Transplant

Microalbuminuria

Mild Retinopathy

Mild Neuropathy

Albuminuria

Macular

Edema

Proliferative

Retinopathy

Periodontal

Disease

Impotence

Gastroparesis

Depression

RISK

Good

PoorSlide5

Administering Insulin

Type I patients must administer insulinInjectionInsulin PumpInsulin PumpAutomatic Detection of Excessive Glycemic

Variability for Diabetes Management Wiley et al.

5Slide6

Continuous Glucose Monitoring

Two approaches to monitoring:FingersticksContinuous Glucose Monitoring (CGM) sensorsCGM Sensor

Fingerstick

Automatic Detection of Excessive

Glycemic

Variability for Diabetes Management Wiley et al.

6Slide7

Data Overload

Automatic Detection of Excessive Glycemic Variability for Diabetes Management Wiley et al.7

CGM sensors record values every 5 or 10 minutesSlide8

Excessive Glycemic Variability

Characterized by fluctuations in blood glucosePatients are not yet routinely screenedAutomatic Detection of Excessive Glycemic Variability for Diabetes Management Wiley et al.8

Acceptable

Glycemic

Variability

Excessive

Glycemic VariabilitySlide9

BackgroundPreliminary work:

Two physicians individually classified 400 plotsNaïve Bayes classifier: 85% accuracyAutomatic Detection of Excessive Glycemic Variability for Diabetes Management Wiley et al.

9Slide10

BackgroundPreliminary work:

Two physicians individually classified 400 plotsNaïve Bayes classifier: 85% accuracyThree orthogonal directions:Smoothing of blood glucose dataFeature engineeringEvaluation of other classifiersAutomatic Detection of Excessive Glycemic Variability for Diabetes Management Wiley et al.

10Slide11

Smoothing Blood Glucose DataSensors record at ±20% accuracy

Physicians implicitly smooth noiseAutomatic Detection of Excessive Glycemic Variability for Diabetes Management Wiley et al.11Slide12

Smoothing Blood Glucose DataSeveral

smoothing methods were investigated:Moving averagesPolynomial regressionDiscrete Fourier transform filterCubic spline interpolationCubic spline identified as the best matchAutomatic Detection of Excessive Glycemic Variability for Diabetes Management Wiley et al.

12Slide13

Cubic Spline Interpolation

Two reasons:Smooth curves Significant pointsDoctorSpline

Automatic Detection of Excessive Glycemic

Variability for Diabetes Management Wiley et al.

13Slide14

Feature EngineeringSeveral features were investigated in this work:

Automatic Detection of Excessive Glycemic Variability for Diabetes Management Wiley et al.14

Mean Amplitude of

Glycemic Excursions (MAGE)Distance Traveled

Excursion Frequency

Standard DeviationArea Under the Curve

Roundness Ratio

Bending

Energy

EccentricityAmplitudes of DFT frequenciesTwo dimensional central momentsDirection CodesSlide15

Feature SelectionTwo methods are reported:

t-Test filterGreedy backward eliminationBoth raw and smooth dataAutomatic Detection of Excessive Glycemic Variability for Diabetes Management Wiley et al.15Slide16

Feature SelectionTwo methods are reported:

t-Test filterGreedy backward eliminationBoth raw and smooth dataOut of the four feature sets selected:No feature appeared in all four setsEccentricity and bending energy were never selectedAutomatic Detection of Excessive Glycemic Variability for Diabetes Management Wiley et al.

16Slide17

Experimental EvaluationThree classifiers compared:

Naïve Bayes (NB)Support Vector Machines (SVM)Multilayer Perceptron (MP)Automatic Detection of Excessive Glycemic Variability for Diabetes Management Wiley et al.

17Slide18

Experimental EvaluationThree classifiers compared:

Naïve Bayes (NB)Support Vector Machines (SVM)Multilayer Perceptron (MP)Evaluated with 10-fold cross validationTuned with development datasetFeatures from feature selectionBoth raw and smooth dataAutomatic Detection of Excessive Glycemic Variability for Diabetes Management Wiley et al.

18Slide19

Visual Overview of Evaluation

Automatic Detection of Excessive Glycemic Variability for Diabetes Management Wiley et al.19Slide20

ResultsNB with raw data: 87.1%

Same settings as preliminary work – the baselinet-test filter:SVM with smoothed data: 92.8%Greedy backward elimination:MP with smoothed data: 93.8%Automatic Detection of Excessive Glycemic Variability for Diabetes Management Wiley et al.

20Slide21

ResultsNB with raw data: 87.1%

Same settings as preliminary work – the baselinet-test filter:SVM with smoothed data: 92.8%Greedy backward elimination:MP with smoothed data: 93.8%Overall:Additional features helpedSmoothed data helpedAutomatic Detection of Excessive Glycemic

Variability for Diabetes Management Wiley et al.

21Slide22

Comparison of ROC Curves for Best Classifiers

Automatic Detection of Excessive Glycemic Variability for Diabetes Management Wiley et al.22Slide23

Future Work This experiment was constrained by the dataset size

Potentially suboptimal parameters and featuresCollecting more data is a high priorityAutomatic Detection of Excessive Glycemic Variability for Diabetes Management Wiley et al.23Slide24

Future Work This experiment was constrained by the dataset size

Potentially suboptimal parameters and featuresCollecting more data is a high priorityPlans for a “5-star” ordinal schemeAlleviate disagreements between annotatorsOpportunity to further improve accuracy Automatic Detection of Excessive Glycemic Variability for Diabetes Management Wiley et al.

24Slide25

From Bench to BedsideIntegration with current CGM management systems

Glucose sensors are in common use as a diagnostic toolSome patients use glucose sensors year roundManufacturers already provide reportingAutomatic Detection of Excessive Glycemic Variability for Diabetes Management Wiley et al.

25Slide26

From Bench to BedsideIntegration with current CGM management systems

Glucose sensors are in common use as a diagnostic toolSome patients use glucose sensors year roundManufacturers already provide reportingDevelopment of a screen for routine clinical useIdentify at risk patients in the clinical settingCompletion of the intended applicationAutomatic Detection of Excessive Glycemic Variability for Diabetes Management Wiley et al.

26Slide27

Acknowledgements

National Science FoundationMedtronicOhio UniversityOur Dedicated Research NursesMy Fellow Graduate Research AssistantsOver 50 Anonymous Patients with Type 1 Diabetes on Insulin Pump TherapyAutomatic Detection of Excessive

Glycemic

Variability for Diabetes Management Wiley et al.

27Slide28

Thanks!Questions?

Automatic Detection of Excessive Glycemic Variability for Diabetes Management Wiley et al.

28