Qualifications And Experience Presentation
1 like1,240 views
The document discusses the author's background and accomplishments in physics and optical coherence tomography (OCT). It then summarizes the author's work developing algorithms to analyze OCT signals and images for non-invasive glucose monitoring. Key contributions included preprocessing algorithms to reduce noise, aggregating 3D OCT images into 1D signals, and using multivariate statistics to understand trends in the data and identify outliers. The goal was to optimize the system performance and mitigate sources of error like motion artifacts.
Qualifications And Experience Presentation
- 2. Background And Accomplishments Grew up on small farm in South Dakota. Developed strong work ethic. 4-H member for 10 years where I performed community service projects (e.g. clean road ditches, cemetery clean-up) and I exhibited livestock, horticulture, baking goods, etc. Physics B.S degree from South Dakota’s premier engineering and science university. Graduated with honors (GPA: 3.5/4.0). Received a minor in mathematics. Prepared and instructed freshman physics recitation courses. Physics M.S. and PhD degree from the University of Tennessee. Performed research at the Center for Laser Applications, University of Tennessee Space Institute. Graduated with honors (GPA: 3.7/4.0). Research and stipend funded with NASA space grant fellowship. Jordan G. Ennis fellowship award that was based on scholarly merit. Outstanding Graduate Research Assistant Finalist Vice President of Finance and Records and Senator in the Student Government Association Hobbies and interests: spending time with family, fishing, camping, hiking
- 3. •Performed independent and collaborative research towards the development of a medical device for non-invasive glucose monitoring using optical coherence tomography (OCT) methods. •Created customized signal and image processing algorithms to analyze and quantify glucose induced scattering changes by utilizing MATLAB toolboxes (e.g. Statistics, Curve Fitting, Signal Processing, Image Processing). •Responsible for analyzing human clinical trial results and afterwards preparing technical reports, summaries, and quantitative analyses for other GlucoLight team members. •Theoretically modeled the medical device imaging performance using ZEMAX software by evaluating the geometric image formation, optical aberrations, stray light analysis, and the optical transfer function. •Experimentally evaluated the medical device imaging performance by constructing an external camera detector system utilizing LabVIEW software control and the Vision Development Module. •Experience working closely with outside testing laboratories and research institutions. •Commanded an extensive knowledge of the relevant scientific literature: glucose monitoring devices, optical and biomechanical skin properties, and OCT principles and applications.
- 4. Optical Coherence Tomography Interferometric technique with a broadband optical source that avoids detection of multiple scattered photons. OCT system’s signal can only form when the optical path length in the sample arm matches that in the reference arm within the coherence length of the source. Key advantage is the capability of detecting photons backscattered from different layers in the sample with high Larin, K., Motamedi, M., Ashitkov, T., and Esenaliev, R., “Specificity of resolution (~10–20 μm). noninvasive blood glucose sensing using optical coherence tomography technique: a pilot study,” Phys Med Biol., 48(10), 1371- 1390 (2003).
- 5. Glucose Measurements with OCT Increasing glucose increases the refractive index of the base medium, and thus decreases the refractive index mismatch between the base medium and the scattering centers. Decreases the scattering of the sample and subsequently lowers the intensity of backscattered photons detected by the OCT instrument. I = I 0 exp( − μt z ) Slope of the Beer-Lambert exponential law is proportional to the total attenuation coefficient of ballistic photons, µt = µa + µs. Since µa << µs in the near-infrared spectral range, the change in the slope is proportional to the change in the scattering coefficient Larin, K., Motamedi, M., Ashitkov, T., and Esenaliev, R., “Specificity of noninvasive blood glucose sensing using optical coherence tomography technique: a pilot study,” Phys Med Biol., 48(10), 1371- 1390 (2003).
- 6. Signal and Image Processing •Developed advanced algorithms to preprocess OCT image. •Filter out noise from 100 200 300 400 500 600 700 800 900 1000 input signals (frequency Axial depth ( μm) 75 filter design, Fourier Stratum corneum analysis) 70 Prickle cell layer •Correlation 65 •Thresholding R fle d p w r (d ) e cte o e B Epidermis Dermis 60 •Aggregate 3-D OCT 55 image into a 1-D 50 exponential signal. 45 40 35 0 200 400 600 800 1000 Axial depth ( μm)
- 7. Algorithm Development and Improvement Analyze slope and morphological p p signal changes associated with glucose changes. Characterize the data by aggregating results from multiple test subjects using MATLAB statistics and curve fitting toolbox. Redesigning algorithms Performed optimization with several dependent variables to 0 50 100 Paramter # 150 200 250 achieve the best overall system performance.
- 8. Multivariate Statistics/Analysis Principal component 3.5 Best diff: 0.0035; Base diff: 0.23; PC diff: 3.6e-005 Best RT data reduction: all Best RT data reduction: subset analysis Baseline RT data reduction: all 3 Baseline RT data reduction: subset PC data reduction: all PC data reduction: subset 2.5 2 Linear regression analysis 1.5 1 Understand trends within 0.5 the data and factors that 0 -1 -0.8 -0.6 -0.4 -0.2 0 Data reduction pearson 0.2 0.4 0.6 0.8 1 change from person to p 0.7 10 person and between 13 0.6 different systems/sensors. 16 0.5 Assisted engineering 0.4 17 y department to identify 4 0.3 system outliers. 8 0.2 0.1 9 10 13 16 17 4 8 9
- 9. 1 0.9 Mitigate 0.8 Sources of 0.7 0.6 Error 0.5 •Quantify and correct 0.4 motion artifacts. 0.3 0.2 •Identify sensor/skin 0.1 interface and general 0 30 35 40 45 50 55 60 65 70 75 loss of power 0.1 problems. 0.09 0.08 •Recognize system 0.07 failures. 0.06 0.05 •Temporal system 0.04 performance degradation 0.03 •Improper sensor 0.02 0.01 placement. 0 30 35 40 45 50 55 60 65 70 75
- 10. Feature Selection and Data Mining Technique to select a subset of relevant features for building robust learning models. Random Forest Support Vector Machines 2 Gtrain13 Btrain13 1.5 Btrain9 Gtrain4 Btrain4 1 Gvalid9 Bvalid9 Gvalid4 Bvalid4 0.5 Train line 0 -0.5 -1 -1.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5 3
- 11. Calibration Analysis Number of independent blood stick measurements to use for calibrating OCT signal into actual glucose concentration. Choosing an appropriate initial calibration gain. Subsequent calibration points Frequency Quality and weight Developing appropriate performance metrics
- 12. Design, Analyze and Present Clinical Trial Results Assisted in designing clinical tests based on results Vary time when meal tolerance glucose test is applied Analyzed hypoglycemic, euglycemic, and hyperglycemic conditions. Physiological lag out of hypoglycemia. Generated an overall and a subject specific report of the clinical trial results for GlucoLight co‐workers using customized MATLAB algorithms and MATLAB Report Generator.
- 13. 1 0.9 0.8 Doppler D g e o p rfu n (a .) 0.7 e re f e sio .u 0.6 OCT: Blood 0.5 0.4 Flow 0.3 0.2 0.1 •Analyzed high and low 0 200 300 400 500 600 700 800 900 1000 frequency components of Axial depth ( μm) 70 OCT interferogram to measure localized perfusion 65 within the dermis due to the Doppler shift caused by the 60 R fle d p w r (d ) e cte o e B moving scatterers (i.e. red 55 blood cells). 50 •Glucose induced scattering 45 changes were measured over perfused and unperfused 40 tissue layers. 35 200 300 400 500 600 700 800 900 1000 Axial depth ( μm)
- 14. 1 OCT Signal Increased 175 B o g co co ce tra n (m /d ) g L Glucose 0.75 levels of O T slo e sig a (a .) n l .u perfusion Correlation lo d lu se n n tio 150 0.5 p 125 C •Observable layers of 0.25 perfusion have high 100 Pearson linear correlation 0 coefficients, 0.94 (420 µm) 0 50 100 150 200 250 300 350 Test time (min.) and 0.91 (680 µm). 1 Immeasurable levels of 175 B o g co co ce tra n (m /d ) perfusion g L 0.75 O T slo e sig a (a .) n l .u lo d lu se n n tio 150 0.5 •Layers with unobservable p perfusion, 0.46 (270 µm) 125 C and 0.61 (490 µm). 0.25 100 0 0 50 100 150 200 250 300 350 Test time (min.)
- 15. Sensor Optical Modeling Modeled OCT imaging sensor with Zemax and varied the optical parameters to measure and optimize physical and optical characteristics: Spot size Raster size RMS Wavefront aberration
- 16. •Constructed a novel electro-optical imaging medical device to detect and autonomously quantify ocular disorders. •Applied real-time adaptive medical device instrument control and data acquisition using LabVIEW software techniques and USB based measurement and automation devices. •Performed image processing software analyses to identify several ocular biometrics using the LabVIEW Vision Development Module and MATLAB Image Processing Toolbox. •Theoretically analyzed the ocular image used to measure and quantify the ocular disorders by incorporating eye models and the device's optical parameters into ZEMAX computer modeling software. •Published in scientific journals and presented oral and poster presentations during scientific conferences/meetings. •Gained experience writing research proposals to government, military, and aerospace external funding agencies.
- 17. Lab Bench Prototype •Independently constructed ocular electro-optical device. •Performed theoretical ZEMAX optical modeling prior to device construction. •Responsible for contacting vendors to purchase optical, electrical, and detector components. •All instrument control and data acquisition was developed using LABVEIW software.
- 18. Strabismus
- 20. Refractive Errors 1.0 •Estimates the retinal reflex by integrating 0.9 portions of the retinal reflection intensity 0.8 from coaxial and eccentric images. 0.7 •1-D and 2-D Gaussian surface fitting is 0.6 performed to estimate the reflex width 0.5 which is related to larger refractive errors. 0.4 0.3 •Integrated intensity ratio method predicts 0.2 the refractive error for smaller refractive 0.1 errors. -40.0 -30.0 -20.0 -10.0 0.0 10.0 20.0 30.0 40.0 Photorefraction Eccentricity (mm)
- 22. Refractive Error 24 FWHM vs. refraction: 5mm pupil diamter Thin lens prediction Results 22 Avg human eye prediction Max FWHM Fitted Gaussian FWHM (mm) Min FWHM 20 Mean FWHM Experimental and theoretical 18 results show that for larger 16 refractive errors the Gaussian 14 FWHM parameter becomes linear 12 with the refractive error. 10 8 Difference between actual 6 -6 -5 -4 -3 -2 -1 0 1 2 3 4 refractive error and experimental Refractive error (D) FWHM prediction was calculated to Intensity ratio vs. refraction: 5mm pupil diamter be less than ~0.7 D. 8 Thin lens prediction Avg human eye prediction Fitted integrated intensity ratio 7 Max ratio Min ratio Difference between the actual 6 Mean ratio refractive error and experimental 5 intensity ratio prediction calculated to be at most ~1 D. 4 3 Cylindrical measurement error was 2 found to be less than 0.6 D. 1 -6 -5 -4 -3 -2 -1 0 1 2 3 4 Refractive error (D)
- 23. High‐Order Aberrations Experimental image data was acquired from human subjects in a clinical environment with different types and amounts of high-order aberrations to determine the capability to differentiate HOAs.
- 24. High‐Order Aberration Results Subject with insignificant high-order aberrations (near-sighted subject). Subject with elevated high-order aberrations (keratoconus subject).
- 25. High Order Aberration Results •Complex reflex intensity patterns were associated with different orders of Zernike polynomials. •Image results from refractive error subjects (N) without HOAs had an average value of 82% of their reflex described with the vertical tilt Zernike term. •Conversely, mild and moderate KC subjects had an average vertical tilt contribution of 31% (A), advanced KC subjects had 44% (B), very mild KC subjects had 79% (C), and subjects after corneal surgery had 34% (D). •Subjects with HOAs tended to have a higher percentage of their reflex images described with higher order Zernike terms.
- 26. Grant Proposals National Institute of Health (R21) Pediatric vision screening (e.g. strabismus, refractive errors). Early keratoconus detection. NSBRI and TATRC Autonomous system to detect vision problems associated with abnormalities and irregularities of corneal and optical opacities resulting from cataracts and foreign objects.
- 27. If you have any questions please don’t hesitate to call or email me. I can be reached via phone at 610‐217‐ 9141(cell) or 931‐841‐3188(home) or via email at dr.kevin.charles.baker@gmail.com.