Optical Coherence Tomography- Signal Processing and Algorithm

OCTDSP for OCT Signal Processing

Applications of OCTGratitude

OPTICAL COHERENCETOMOGRAPHY:SIGNAL PROCESSING AND

ALGORITHM

OPTICAL COHERENCE TOMOGRAPHY



What is OCT?Principle and InstrumentationOCT SystemsOCT Systems ProcessingTypes of OCT SystemsBasic Signal Processing Chain in OCT Systems

What is OCT?

Medical imaging modality with 1-10 µ m resolutions and1-2 mm penetration depthsHigh-resolution, sub-surface non-invasive or minimallyinvasive internal body imaging technique for structural andquantitative imagingSignal processing intensive suited for embeddedimplementations using digital signal processors (DSP) andsystem-on-chip (SoC) application processorsPerformance given in cycles per scanline and total numberof scanlines that can be processed per secondEnables cross-sectional imaging of tissue microstructure insitu and in real time





Principle and Instrumentation:

Based on the principle of low coherence interferometryBy stacking the axial scans in X and/or Y directions,two orthree dimensional imaging.Imaging is performed by measuring the echo time delayand intensity of back-reflected or backscattered light

Performed fiber-optically using delivery devices such ashandheldprobes,endoscopes,catheters,laparoscopes, andneedlesMeasurements are performed using a Michelsoninterferometer with a low coherence length light source





OCT Systems Using Michelson Interferometer

A standard MichelsonInterferometer with alow-coherence light source isusedIncoming broadband beam oflight is split into the referencepath and the sample pathAfter back-reflection from thereference mirror and the multiplelayers of thesample,respectively,they arerecombined





OCT Systems Using Michelson Interferometer

Thus an interference signal is formedDue to broadband nature of light interference of the opticalfields occur only when the path lengths of the referenceand the sample arm are matched to within the coherencelength of the lightThis interference signal carries information about thesample at a depth determined by the reference path length

SHAKIRA THANKAYATHIL OPTICAL COHERENCE TOMOGRAPHY




Steps in OCT System processing

Signal processing intensiveReal-time data is to be acquiredAcquired data is processed to extract meaningfulinformationand then the information is displayed





Time-domain OCT(TD-OCT)

Sample arm ofthe interferometerilluminates thelight on the tissueand collects thebackscatteredlight





Time-domain OCT(TD-OCT)

Reference arm of the interferometer has a reference pathdelay which is scanned as a function of timeOptical interference between the light from the sample andreference arms occurs only when the optical delays matchto within the coherence length of the light sourcePhoto-detector detects the average intensity over therange of frequenciesThe detected signal consists of a DC term and aninterference term that contains the sample informationDual balanced approach is used to subtract a portion ofsource signal through the use of a second photo-detectorbefore digitizing the signal is used to remove the DC termpartially





Spectral-domain OCT(SD-OCT)

A broadband-source oflight with short temporalcoherence length is usedas an input to theinterferometer.Reference mirror positionis fixed





Spectral-domain OCT(SD-OCT)

Echoes of light are obtained by Fourier transforming theinterference spectrumDepth information is obtained by measuring the spectraldensity in the detection arm using a spectrometer, wherethe interference beam is dispersed by a diffraction gratingand the individual wavelength components are detected byan array detectorthe path difference remains fixed and is assumed to bezero without loss of generality





Advantage of SD-OCT over TD-OCT

Permits faster acquisition of the entire depthprofile(A-scans)Video-rate imaging is possibleHigh-speed acquisition without any moving partsminimizes any distortion in the OCT images due to motionin the sampleTheoretical signal-to-noise (SNR)ratio is independent ofthe spectral bandwidth of the light source leading toincrease in axial resolution





Swept Source Systems(SS-OCT)

A frequency-swept laseror a tunable laser withjust a single detector isused to obtain thespectrogram






Require rapid tunable, narrow line-width lasers, which usehigh-speed analog-to-digital(A/D) converters andsingle-point detectorsThe sample of received spectrum is probed with narrowband but frequency varying sourceAlso the sample is probed with chirp like signal source andthe received signal is reflectionsfrom the reference andfrom the samplesAfter data capturing the operations are same as SD-OCT






A dual-balanced detection is used to remove the DC beforedigitizationThus full dynamic range can be used to capture theinterference signal





Background Subtraction

To eliminate the reference power term, the referencespectrum from only the reference arm is detected andsubtracted from the interference spectrumThe reference spectrum is acquired at the beginning ofevery image acquisition to account for fluctuations in thesource between measurementsFrom the acquired data derive the reference spectrumsince the interference is usually high frequency fringes,whereas, the background term has low frequencycomponentsIn swept source systems, using dual-balancedphoto-detectors allows this subtraction in analog domain





Re-Sampling

In SD-OCT systems, spectrometers measure opticalintensity as a function of wavelengthIn order to apply the fast Fourier transform (FFT)reconstructing the axial scan as a function of depth, thespectrum should be evenly sampled in k-spaceTherefore, the spectrometer output must be transformedfrom the wavelength to the frequency space





Image Formation (FFT)

The basic operation to get the depth resolved A-scan fromthe interference fringesThe structural image is obtained by taking the magnitudeof the complex FFT outputEach FFT creates a particular A-scanBy moving the galvanometer in x direction,the successiveA-scan line is created, which is stacked together to createa B mode imageBy moving the galvanometer in both x-y direction, a full 3Dvolume can be generated





Display

Two dimensional OCT images are typically representedusing a density plotThe horizontal axis typically corresponds to the direction oftransverse scanning and the vertical axis corresponds tothe scanning depthA gray level is plotted at a particular pixel on an imagecorresponding to the magnitude of the depth profile at aparticular depth and transverse scanning position.Pixel intensity range is compressed using the logarithmicnon-linearity before displaying it





Image Enhancement

Speckle noise that arises from the interference betweencoherent waves backscattered from nearby scatters in asample is the dominating source of noise in OCT imagesNon-linear direction preserving digital filters such as meanand median filters are used to improve the image qualitySimple signal averaging over the same line can also beused to improve the signal-to-noise ratio of the datacollected at the cost of reduced frame rateA secondary camera is sometimes used to trackinvoluntary movements and control the data acquisition ina closed loop manner





Dispersion Compensation

The refractive index of the biological tissues is, in general,frequency dependent slowing down certain opticalfrequencies to a greater extent than others, therefore,dispersing the lightDispersion correction can take place both in the hardwareand the softwareAs the sample being imaged itself could also bedispersive,an automated numerical method of dispersioncompensation is desirable




OCT Algorithm Implementation on DSP

DSP

Benchmarking of the key signal processing algorithmsneeded to produce a B-mode image was done on TexasInstruments’ C64x+ DSP architectureTI has several variations of processors based on C64x+architectureThese devices allow development of OCT systems at afractional power compared to x86 or graphics processingunit (GPU) based units while maintaining the sameprogrammability feature





DSP

Within limits of processing capabilities, the same devicecan be used to perform various modes of operations likeB-mode imaging, Doppler, polarization sensitive,etc.Due to programmability, the same processing unit used formain signal chain can be utilized for calibration anddifferent estimation algorithms needed to identify systemparametersDue to smaller footprint, DSP-based systems allow thedevelopment of smaller, low power, low cost as well asbattery-operated portable systems





Basic Introduction on OCT Algorithms

Using TI’s Embedded Processor Software Toolkit forMedical Imaging provides optimized functions toimplement such a signal processing chain on DSPs basedon the C64x+ architectureTI has several variations of processors based on C64x+architectureFigure shows the basic signal processing chain needed tocreate a structural image from the recorded interferencesignal





Background Subtraction

A simple operation where the background is subtractedfrom the acquired dataThis subtraction takes care of the DC component in thesignal that is due to the reflectance from the reference armAdditional variations due to fixed pattern noise in the linescan camera and variations in power spectral densities ofthe source can also be suppressed by this method





Resampling

In SS-OCT, the frequency sweeping is usually non-linear infrequency (or k-space)In SD-OCT, spectrometers are used that measure opticalintensity as a function of wavelength.The signal obtained isnon-equidistant in frequency (or k) space





Resampling

In either of the systems, the captured information is notlinearly spaced in frequencyA re-sampling technique is usually employed to resamplethe recorded discrete intensities from the acquired domainto linear frequency domainThe cubic spline interpolation algorithm is used as definedin to perform the re-sampling function





FFT

After resampling, the data is linear in the k-space and anFFT is performed to reconstruct the axial scan as afunction of depthThe MED-STK has several variations of the FFT availablefrom the TI DSPLibThe routines use 32 bit internal precision math operationswith 16-bit twiddle factors and only allows power of 2 FFTsizesFor the sizes of FFT used in OCT with 16-bit input, there isno possibility of internal saturation





FFT

The 16-bit output is derived using your programmable rightshift value

Magnitude ComputationLog Compression





API

A DSP-based software implementation of these OCTalgorithms consists of well defined APIsFor all the OCT algorithms through the API all the physicalparameters related to the image of interest such asnumber of scanlines, number of samplers per scanline,number of frames, etc should be specifiedSpecific algorithms that have additional parameters shouldalso be specified





Flexibility

The DSP-based OCT algorithms should be flexible interms of having the ability to operate in different modesThe same DSP used for the main signal chain can beutilized for calibration and different estimation algorithmsneeded to identify system parameters like backgroundsignal, the re-sampling points, the phase corrections fordispersion compensation, etcThese parameters are either pre-computed duringcalibration or computed automatically before the imageacquisition process.





Efficiency

The implementation should be highly efficient so thatminimum DSP CPU bandwidth is consumed for thesealgorithms, allowing more space for future OCT algorithmsto be implemented





I/O Bandwidth Requirement

The I/O bandwidth requirements for CPUs to access all thenecessary data is algorithm dependent to a certain extentIn the resampling algorithms, eight lines of data have to beaccessed simultaneously for the efficient implementation ofthe algorithmThe effect of cache on benchmarking depends on memoryorganization and system function partitioningAll of the benchmarking provided here is done on aTMS320C6455 device





Precesion Requirement

The system will suffer from poor images if the precisionsused throughout the system are not sucffientIf more than necessary precision is used,that willunnecessarily increase the cycle count and reduce thenumber of scanlines that can be processed through thesystemThe APIs for 16-bit input and output have been optimizedexcept for the compression module, which outputs 8-bitdata for visualizationSince the analog to digital converter samples theinterference data with 10-14 bit precision,16 bit is sucientfor high-quality image production for this application




Clinical Applications

As a noninvasive biomedical imaging modality that enablescross-sectional visualization of tissue microstructures invivo.Architectural morphology to be visualized in situ and in realtimeEnables imaging of structures in which biopsy would behazardous or impossible, and promise to reduce thesampling errors associated with excisional biopsy




Clinical Applications

Translated from bench to various clinical applicationsincluding ophthalmology , cardiology , gastroenterology ,dermatology , dentistry , urology, gynecology , amongothersDue to programmability, the same processing unit used formain signal chain can be utilized for calibration anddifferent estimation algorithms needed to identify systemparametersThe most developed clinical OCT applications are thosefocusing on ophthalmic, cardiovascular,and oncologicimaging




THANK YOU

Mr.VINOD G.Mr.HARIKRISHNAN A.I.


Optical Coherence Tomography- Signal Processing and Algorithm

Documents