Bio-optical Sensing Dissertation
-
Upload
ricky-hennessy -
Category
Healthcare
-
view
141 -
download
0
Transcript of Bio-optical Sensing Dissertation
Title Slide
Depth Resolved Diffuse Reflectance SpectroscopyRicky Hennessy
The Biomedical Informatics Lab (BMIL)The Biophotonics Laboratory
UT Biomedical Informatics Lab
UT Biomedical Informatics Lab0
Committee1/60
Mia K. Markey, Ph.D. - AdvisorThe Biomedical Informatics Lab
James W. Tunnell, Ph.D. - AdvisorThe Biophotonics Lab
Stanislav Emelianov, Ph.D.Ultrasound Imaging and Therapeutics
Andrew K. Dunn, Ph.D.Functional Optical Imaging Lab
Ammar M. Ahmed, M.D.Dermatology at Seton
UT Biomedical Informatics LabDiffuse Reflectance Spectroscopy (DRS)2/60
UT Biomedical Informatics LabApplications of DRS3
Cancer Detection
Soil CharacterizationWearable Tech
Food Quality
Endoscopic Surgery
Cosmetic Applications
UT Biomedical Informatics LabCancer Detection with DRS4/60
Extract Features from DataUse Features to Create ClassifierRajaram et al. Lasers Surg Med 42:876-887 (2010)
UT Biomedical Informatics LabDRS Instrumentation5
SpectrometerLight SourceSource FiberDetector FiberComputerFiber Bundle
UT Biomedical Informatics LabLight Absorption in Tissue6/60
Other AbsorbersWaterBilirubinLipidProteinCollagen
UT Biomedical Informatics LabLight Scattering in Tissue7/60
UT Biomedical Informatics LabBiological Origins of Scattering8/60
Scattering is caused by index of refraction mismatches
Cells = ~10 mNuclei = ~1 mCollagen = 0.1 mMembranes = 0.01 m
UT Biomedical Informatics LabScattering Coefficient (s)9/60Scattering coefficient (s) is proportional to concentration of scatterers in a medium
s-1 is the average distance a photon travels between scattering eventss
UT Biomedical Informatics Lab
Direction of Scattering10/60
Isotropic Scattering
Anisotropic ScatteringTissue scattering is in forward direction(g = ~0.9)
Henyey-Greenstein Phase Functiong = 0
UT Biomedical Informatics LabRadiative Transport Equation (RTE)11/60
Energy
Scattering inScattering out
Absorption
UT Biomedical Informatics Lab
Reduced Scattering Coefficient12/60
12109876543
Using reduced scattering with isotropic scattering is equivalent to larger scattering with anisotropic scattering
UT Biomedical Informatics Lab
Diffusion Approximation to the RTE13/60
Kienle et al., JOSA A, 1997, 14(1), 246-254Assumes isotropic scatteringScattering >> absorptionSource Detector Separation > ~ 1 mmBlood is highly absorbingEpidermis is ~100 m thick
UT Biomedical Informatics LabMonte Carlo Simulation14/60
UT Biomedical Informatics Lab15
UT Biomedical Informatics Lab16
UT Biomedical Informatics LabThe Monte Carlo Lookup Table (MCLUT) Method
Hennessy et al., JBO, 2013, 18(3), 037003
UT Biomedical Informatics Lab
UT Biomedical Informatics Lab17
Tissue Properties Optical Properties
SCATTEREDABSORBED
TISSUE PROPERTIES OPTICAL PROPERTIES SIGNALFORWARD MODELSCATTERING PROPERTIES
ABSORBER CONCENTRATIONSHbO2HbMelanin
18/60
UT Biomedical Informatics Lab18
Forward Model Flowchart
TISSUE PROPERTIES OPTICAL PROPERTIES SIGNALFORWARD MODEL
Light Transport ModelScattering at 0, Concentration of chromophoresCalculate absorption and scattering coefficients at each wavelengthScatteringAbsorptionUse light transport model to calculate diffuse reflectanceGenerate diffuse reflectance spectrum19/60
UT Biomedical Informatics Lab19
Monte Carlo on GPU
Modern processor w/ 4 cores = 4 times speedup
Modern GPU w/ 500 cores = 500 times speedup!< $300
Alerstam et al., Biomed. Opt. Express., 2010, 1, 658-67520/60
UT Biomedical Informatics Lab20
----- Meeting Notes (2/1/14 10:54) -----Monte Carlo simulation of photon migration in turbid media is a highly parallelable problem, where a large number of photons are propagated independently, but according to identical rules and different random number sequences.
Monte Carlo Lookup Table (1 Layer)
21/60
UT Biomedical Informatics Lab21
----- Meeting Notes (2/1/14 10:54) -----One layer model. LUT created using GPU MCML. Sample modeled spectrum with melanin and hemoglobin
MCLUT Inverse ModelREFLECTANCE OPTICAL PROPERTIES TISSUE PROPERTIES
OptimizationRoutine22/60
UT Biomedical Informatics Lab22
----- Meeting Notes (2/1/14 10:54) -----Inverse model. Just explain the flow chart
CalibrationMCLUT Modeled SpectraRMC = photons countedMeasured SpectraRmeas = Iraw/Istandard
Modeled spectra and measured spectra have same optical properties
23/60
UT Biomedical Informatics Lab23
Validation of One-Layer Model
RMSPE = 2.42%RMSPE = 1.74%Validation with 3 X 6 matrix of phantoms containing hemoglobin and polystyrene beads.Decreased percent error of 3.16% and 10.86% for s' and a, respectively, when compared to experimental LUT method24/60
UT Biomedical Informatics Lab24
Errors Caused by One-Layer Assumption for Skin
Hennessy et al., JBO, 2015, 20(2), 027001
UT Biomedical Informatics Lab
UT Biomedical Informatics Lab25
Fit Two-Layer Data with One Layer Model26/60
Hb + HbO2melanin
melanin +Hb + HbO21. Create two-layer spectra2. Fit with one-layer model[mel][Hb]SO2ScatteringEpidermal thickness[mel][Hb]SO2ScatteringVessel radiusNotice that the fit is very good
UT Biomedical Informatics LabMelanin27/60
One-Layer model underestimates [mel]Magnitude of error is dependent on epidermal thickness (Z0)Z0 Error
UT Biomedical Informatics LabHemoglobin28/60
One-Layer model underestimates [Hb]Magnitude of error is dependent on epidermal thickness (Z0)Z0 Error
UT Biomedical Informatics LabOxygen Saturation29/60
One-Layer model overestimates SO2 when SO2 < 50%One-Layer model underestimates SO2 when SO2 > 50%Magnitude of error is dependent on epidermal thickness (Z0)Z0 Error
UT Biomedical Informatics LabPigment Packaging30/60
In tissue, blood is confined to vesselsThis significantly reduces the optical path length where absorption is high (Soret Band)Causes a flattening of the absorption spectrum
UT Biomedical Informatics LabVessel Radius vs. Epidermal Thickness31/60
Vessel radius (pigment packaging) factor is highly correlated with top-layer thicknessPigment packaging factor is likely a combination of vessel packaging and epidermal thickness
UT Biomedical Informatics LabCorrelation Between [Hb] and [mel]32/60
R = 0.04R = 0.80One-layer assumption causes artificial correlation between [Hb] and [mel]
UT Biomedical Informatics LabConclusions about One-Layer ErrorsCauses underestimation of [Hb] and [mel]Magnitude of error is function of epidermal thicknessCauses error in SO2 that is a function of epidermal thickness as well as SO2Vessel Packaging factor and epidermal thickness are highly correlatedCauses an artificial correlation in [Hb] and [mel]33/60
UT Biomedical Informatics LabThe Two-Layer Monte Carlo Lookup Table Method
Sharma, Hennessy et al., Biomed Optics Express, 2014, 5(1), 40-53
UT Biomedical Informatics Lab
UT Biomedical Informatics Lab34
Motivation of Two-Layer Model
EpidermisDermisStratum Corneum
Melanin
Hemoglobin35/60
UT Biomedical Informatics Lab35
Motivation of Two-Layer ModelPigmentary disorder studiesDisease (rosacea, lupus, scleroderma, morphea, lymphederma) monitoringTreatment outcome measures for many cosmetic proceduresTopical medical absorption studiesMeasuring thickness of psoriatic plaqueDetermination of epidermal thickness
MelasmaMethod of melasma treatment depends on depth of melanin
Woods lamp is current method to determine location of melaninQualitative More contrast for epidermal melasma. Doesnt work for patients with dark skin.Asawansa et al., Int. J. Derm, 1999, 38, 801-80736/60
UT Biomedical Informatics Lab36
Two Layer MCLUTWe can create a 5D LUTTop layer absorptionBottom layer absorptionScatteringTop layer thicknessSDS37/60
Segment of 5D lookup tableZ0 = 200 mSDS = 200 mR1 = R2 = 100 m
Sharma, Hennessy et al., Biomed. Opt. Express, 2014, 5(1), 40-53
UT Biomedical Informatics Lab37
Two-Layer Forward Model38/60
UT Biomedical Informatics Lab38
Two-Layer MCLUT Inverse Model
39/60
UT Biomedical Informatics Lab39
Two-Layer Phantom Construction
40/60
UT Biomedical Informatics Lab40
Phantom Validation StudyPHANTOMs (mm-1)a,t (mm-1)a,b (mm-1)11.50.251.27521.50.251.27531.52.30.2541.52.30.2551.50.252.361.50.252.372.850.252.382.850.252.392.852.30.25102.851.2750.25112.851.2752.3120.750.251.275130.751.2750.25140.750.252.3
Set of Two Layer PhantomsRange of scattering and absorption in skin
s = 1-3 mm-1a = 0 2.3 mm-1
Nichols et al., JBO, 2012, 17(5), 05700141/60
UT Biomedical Informatics Lab41
Phantom Data42/60
UT Biomedical Informatics Lab42
Results: Top Layer Thickness43/60
Error = 10% for Z0 < 500 m
Photons are not sampling this deep.
UT Biomedical Informatics LabResults: All Optical Properties44/60
UT Biomedical Informatics LabResults: Dependence on SDS45/60
Main TakeawayAccuracy of extracted parameters is dependent on probe geometry. This is due to sampling depth of probe.
UT Biomedical Informatics LabSampling Depth of Diffuse Reflectance Spectroscopy Probes
Hennessy et al., JBO, 2014, 19(10), 107002
UT Biomedical Informatics Lab
UT Biomedical Informatics Lab46
Probe Geometry and Sampling Depth47
TissueSourceFiberDetectionFiberSDS
SamplingDepthZ(a,s,rS,rD,SDS)
rSrD a,s
rS,rD
SDS)
-Tissue Optical Properties-Source/Detector Fiber Sizes-Source/Detector Separation* Credit to Will Goth for this slide47/60
UT Biomedical Informatics LabDefining Sampling Depth48/60
This experiment was performed computationally (MC simulation) and experimentally (phantoms)
UT Biomedical Informatics LabExperimental Validation49/60
E = 1.71%E = 1.27%E = 1.24%SDS = 370 mSDS = 740 mSDS = 1110 mLook at axes to see deeper sampling for larger SDSs
UT Biomedical Informatics LabAnalytical Model of Sampling Depth50/60
This expression can be used to aid in the design of application specific DRS probesE = 2.89%Expression was found using TableCurve 3D. Free parameters [a1, a2, a3, a4] were selected using a least-squares fitting algorithm.
UT Biomedical Informatics LabChoice of g and Phase Function51/60
The choice for g and phase function had negligible impact on the sampling depth model
UT Biomedical Informatics LabApplying the Sampling Depth Model52/60
6-around-1 adjacent fiber orientation with medium (series 1), high (series 2), and low (series 3) absorption.Sampling depth changes with wavelength
UT Biomedical Informatics LabPilot Study: Two-Layer MCLUT Method Applied to In Vivo Data
UT Biomedical Informatics Lab
UT Biomedical Informatics Lab53
Study Population and Data54/6080 SubjectsIRB approval from UT Austin - #0000203051 males, 29 FemalesAverage age of 25.7 yearsAges 18-46
Measured spectra from the following anatomical locationsBackCalfCheekForearmPalm
Measured the followingMelaninHemoglobinScattering Epidermal Thickness
This study is still unpublished
UT Biomedical Informatics LabInstrumentation55/60
x 2Source Diameter = 40 mRing 1 Diameter = 40 mRing 2 Diameter = 200 mRing 1 SDS = 55 mRing 2 SDS = 205 m
Unfortunately, data from ring 1 was unusable
UT Biomedical Informatics LabMelanin56/60
Palm has less melanin, which agrees with the expected result.
Average of 1.83 mg/ml is within range of published values for melanin concentration[0-5 mg/ml]
UT Biomedical Informatics LabHemoglobin57/60
Higher levels of hemoglobin in the face and forearm agrees with the expected results.
Average of 1.37 mg/ml is within range of published values for [Hb][0.5-10 mg/ml]
UT Biomedical Informatics LabScattering58/60
No significant difference between anatomical locations.
Average of 22.75 cm-1 is within range of published values for scattering at 630 nm[15 25 cm-1]
UT Biomedical Informatics LabEpidermal Thickness59/60
No significant difference between anatomical locations.
Average of 90 m is within range of published values for epidermal thickness.[40 200 m]
We expected to see a difference between anatomical locations.
UT Biomedical Informatics LabConclusions about Pilot StudyA two-layer model can be used to extract depth dependent properties from in vivo DRS dataThe results agree with previously published valuesHowever, we expected to see a difference between anatomical locations for epidermal thicknessThis could be due to the absence of data from the inner ring of fibers.60/60
UT Biomedical Informatics LabConclusions about Pilot StudyData should be recollected with multiple SDSsAdditional patient data such as race/ethnicity and skin color should be documented61
UT Biomedical Informatics LabOverall ConclusionsThe MCLUT method is an accurate and fast way to analyze DRS dataA one-layer assumption for skin causes significant errors in DRS data analysisThe MCLUT method can be extended to two-layers, allowing the extraction of depth dependent propertiesDepth sampling of DRS probes can be tuned by changing the probe geometryDRS can be used to measure the depth dependent properties of skin in vivo
62/60
UT Biomedical Informatics Lab
Contributions to Field63The MCLUT Method
Two-Layer MCLUT Method
One-Layer Errors Analysis
DRS Sampling Depth Analysis
UT Biomedical Informatics LabFundingThe NIH (R21EB0115892)The NSF (DGE-1110007)CPRIT (RP130702)64
UT Biomedical Informatics LabAcknowledgements65The Biophotonics LabJames W. TunnellWill GothBin YangManu SharmaSam Lim Sheldon BishXu FengVarun PataniThe Biomedical Informatics LabMia K. MarkeyNishant VermaGezheng WenClement SunNisha KumaraswamyHans HuangJuhun LeeGautam MuralidharDaifeng Wang
UT BME StaffMargo CousinsBrittain SobeyMichael Don
UT Biomedical Informatics Lab