A semi-automated analysis method of small sensory nerve fibers in human skin-biopsies

A semi-automated analysis method of small sensory nerve fibers in human skin-biopsies
of 18
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
  A semi-automated analysis method of small sensory nerve fibers in humanskin-biopsies Kazuyuki Tamura a, ∗ , Violet A. Mager a,b , Lindsey A. Burnett c , John H. Olson d , Jeremy B. Brower e,f  , Ashley R.Casano f  , Debra P. Baluch d , Jerome H. Targovnik  g,d , Rogier A. Windhorst h,a , Richard M. Herman f,e a  Department of Physics, Arizona State University, Tempe, Arizona, 85287-1504, USA b Carnegie Observatories, Pasadena, California, 91101-1232, USA c  Molecular and Cellular Biology Program, Arizona State UniversityTempe, Arizona, 85287-4601, USA d  School of Life Sciences, Arizona State University, Tempe, Arizona, 85287-4501, USA e  Harrington Department of Bioengineering, Arizona State University,Tempe, Arizona, 85287-9709, USA  f  Clinical Neurobiology and Bioengineering Research Center, Banner Good Samaritan Medical Center,Phoenix, Arizona, 85006-2666, USA g Carl T. Hayden VA Medical Center, Phoenix, Arizona, 85012-1839, USA h School of Earth and Space Exploration, Arizona State University,Tempe, Arizona, 85287-1404, USA Abstract ComputerizedDetection Method (CDM) software programs have been extensively developed in the field of astron-omy to process and analyze images from nearby bright stars to tiny galaxies at the edge of the Universe. These object-recognition algorithms have potentially broader applications, including the detection and quantification of cutaneousSmall Sensory Nerve Fibers (SSNFs) found in the dermal and epidermal layers, and in the intervening basementmembrane of a skin punch biopsy. Here, we report the use of astronomical software adapted as a semi-automatedmethod to perform density measurements of SSNFs in skin-biopsies imaged by Laser Scanning Confocal Microscopy(LSCM). In the first half of the paper, we present a detailed description of how the CDM is applied to analyze theimages of skin punch biopsies. We compare the CDM results to the visual classification results in the second half of the paper. Abbreviationsused in the paper, descriptionof eachastronomicaltools, and theirbasic settings and how-tosare described in the appendices. Comparison between the normalized CDM and the visual classification results onidentical images demonstrates that the two density measurements are comparable. The CDM thereforecan be used —at a relatively low cost — as a quick ( a few hours for entire processing of a single biopsy with 8-10 scans ) and reliable( high-repeatability with minimum user-dependence ) method to determine the densities of SSNFs. Key words:  Computerized detection method, Visual classification, Small sensory nerve fibers, Basement membrane,Confocal images ∗ Corresponding author. Tel.:  + 1-480-965-0665; fax:  + 1-480-965-7954.  Email addresses:  ktamura@asu.edu  (Kazuyuki Tamura), vmager@ociw.edu  (Violet A. Mager), lindsey.burnett@asu.edu  (Lindsey A. Burnett), john.olson@asu.edu  (John H. Olson), jeremy.brower@asu.edu  (Jeremy B. Brower), ashley.casano@asu.edu  (Ashley R. Casano), page.baluch@asu.edu  (Debra P. Baluch),  docglands@cox.net (Jerome H. Targovnik),  rogier.windhorst@asu.edu  (Rogier A.Windhorst),  richard.herman@cox.net  (Richard M. Herman) 1. Introduction Damageto the peripheralnervoussystem can beidio-pathic, or can occuras a result of known causes, such asof pre- and post-diabetic disorder (e.g., Lacomis, 2002;Smith and Singleton, 2006). This damage can be char-acterized by pathology of Small Sensory Nerve Fibers(SSNFs: e.g., Dyck et al., 1981; Periquet et al., 1999;Verghese et al., 2001; Sumner et al., 2003; Malik et al.,2005; Smith and Singleton, 2006), which are di ffi cult toassess functionally, and may be variable in symptoms Preprint submitted to Journal of Neuroscience Methods October 9, 2009  (e.g., pain and tingle), signs (e.g., sensory loss), or dis-abilities.The di ffi culty in diagnosing an SSNF neuropathyhasled to the development of histological analysis of skinpunch biopsy tissue, utilizing morphological surrogatemarkersandconfocal,orconventionalmicroscopy(e.g.,Dalsgaard et al., 1989; Karanth et al., 1991; Kennedyand Wendelschafer-Crabb, 1993). This technique is asafe, reliable, and reproducible method of quantifyingSSNF pathology. It also has been advocated for di-agnosing SSNF neuropathy from a variety of causes(Holland et al., 1997, 1998), and for assessing the de-gree of neuropathyfrom none to severe (Quattrini et al.,2007). A basis for such diagnostic procedure is thatthese biopsy tissues are rich in SSNFs (often referredto as C-fibers), that can be labeled by immunostainingwith a pan-neuronalmarker(e.g., Protein Gene-Product(PGP) 9.5; Dalsgaard et al., 1989; Karanth et al., 1991),and quantified as density values (e.g., fibers mm − 1 ) of epidermallength, or as in this presentation,of basementmembrane length where fibers are counted in the base-ment membrane.Under normal clinical conditions, the key issue in vi-sually measuring the density of SSNFs is the reliabil-ity. The use of multiple blind-study classifiers is usu-ally used to improve the reliability of the analysis. Sta-tistical studies show the correlation coe ffi cients for re-peated observations by the same classifier is 0 . 80 ± 0 . 06(McArthur et al., 1998), and for observations betweendi ff  erent classifiers is 0 . 90 ± 0 . 04 (Hirai et al., 2000).Smithet al.(2005)showsthatthe interobservervariabil-ity is 9.6% ± 9.4 (in the form of relative intertrial vari-ability (RIV)  ±  standard deviation (SD)), and intraob-server variability is 9.6% ± 8.9 for each biopsy, wherea RIV value of less than 10% indicates a high degreeof repeatability. Detail descriptions of how to calculatethe RIV and SD are provided by Smith et al. (2005). Toimprove the reliability, a number of computerized pro-grams have been adopted (e.g., Kennedy et al., 1996).One practical issue with these programs is the cost of the software. We therefore propose here: (1) a  low-cost  , novel, and semi-automated computerized detec-tion method, which can determine the density of SS-NFs in immunostained tissue images; and (2) that thismethod, with some adaptation, can be readily used todetect SSNFs, and to determine the presence of SSNFneuropathyin any kind of tissue.In this paper, we present a quantitative semi-automated method of measuring the SSNF densitieswithin the basement membrane, using parts of exist-ing computer codes. These codes have been success-fully used for decades to analyze images of tiny galax-ies in the field of extragalactic astronomy (e.g., Driveret al., 1995; Odewahn et al., 1996; Arnouts et al.,1997). Since the 1960’s, astronomers have developedmany techniques for detection of those tiny objects.As a result, current technology allows us to detect andanalyze from the surface of the Sun to objects 10 22 times fainter than the Sun, and up to 10 million timessmaller than the angular size of the full Moon (e.g.,Tyson, 1988; Neuschafer and Windhorst, 1995; Pas-carelle et al., 1996; Windhorst et al., 1998, 2008). As-tronomers also have developed numerical techniques toobtain the best possible resolution through deconvolu-tion and other methods such as the Maximum EntropyMethod(MEM: Gull and Daniell, 1978,1979),CLEAN(H¨ogbom, 1974), and Pixon methods (Pina and Puetter,1993; Puetter, 1997). Some of these (e.g., MEM andPixon methods) have been adapted for use in the med-ical field, such as the detection of breast cancer fromX-ray mammograms. For these reasons, we employthe rich expertise of astronomical image analysis soft-wareto detectSSNFs in thebasementmembraneofskinpunch biopsies of volunteer subjects. 2. Materials and Methods All human studies have been approved by the In-stitutional Review Boards of two institutions: BannerGood Samaritan Medical Center, Phoenix, AZ and Ari-zona State University, Tempe, AZ. They have been per-formed in accordance with the ethical standards laiddown in the Declaration of Helsinki. All persons gavetheir informed consent prior to the investigation. Agroup of 27 subjects were recruited yielding 52 biop-sies. Skin punch biopsies were obtained from the prox-imal forearm and  /  or distal thigh of these subjects.The intent of this work is to demonstrate that SSNFdensity values attained by computer analysis com-paresfavorablytothevisually-monitoreddensitycountswithin the same subject. Thus, in any one subject, ahealthy or unhealthy pool of subjects is not specificallyrelevant to the research question. Nevertheless, in thisstudy, we chose to examine SSNF densities in a repre-sentative cohort of obese subjects with no neuropathicsymptoms with the purpose to conduct a later study,which would examine the present cohort with anotherobese population with diabetes. 2.1. Sample preparation: Procedures2.1.1. Biopsy technique A 2cm circle was drawn around the hairy skin sitesof the proximo-lateral forearm and the disto-medial as-2  pect thigh. Each skin site was then sterilized with an al-cohol swab and anesthetized with 1% lidocaine. Oncethe subject reported complete numbness in the area, askin punch biopsy was performed (3mm diameter by3mm depth). The sample was placed in fixative, anda small sample of gelfoam was placed on top of thewound to expedite healing. Bacitracin  R  (a topical an-tibiotic) was swabbed on a bandage and placed on topof the site. There were no reports of adverse events as aconsequence of these procedures. 2.1.2. Sectioning and staining Skin punch biopsies were fixed overnight in modi-fied Zamboni’s fixative (2% formaldehyde,15% Picricacid in 0.1M PBS, pH 7.5); biopsies were equilibratedin 50% sucrose in 0.01M PBS for 12 hours at 4 ◦ C,andmountedinTissueTekOCTCompound(MilesInc.,Elkhart, IN); 0.06mm (60  µ m) thick sections were ob-tained by cryostat sectioning. Biopsy sections werethen blocked in blocking solution (0.1M PBS, 0.05%Tween-20, 0.1% Triton X-100, 5% bovine serum al-bumin frac V) at 4 ◦ C for 12 hours, then washed twiceby solution replacement with washing solution (0.1MPBS, 0.05% Tween-20, 2% BSA, pH 7.5). Primary an-tibodies to collagen IV (mouse anti-human collagen IVmonoclonal,Cat. #MAB1910, Chemicon International,Temecula, CA) and PGP 9.5 (rabbit anti-human PGP9.5 IgG purified polyclonal, Cat. #7863-0507,BioGen-esis, Kingston, NH) were diluted 1:50 with antibodydilution bu ff  er (0.1M PBS, 1% BSA, pH 7.5). Sec-tions were immunolabeled for 24 hours at 4 ◦ C in theprimary antibody solution. Sections were then washedtwice in washing solution. Secondaryantibodies (AlexaFluor 488 goat anti-mouse IgG (H + L); Cat. #A11001,and Alexa Fluor 633 goat anti-rabbit IgG (H + L), Cat.#A21070,MolecularProbes, Eugene,OR) were diluted1:100 in antibody dilution bu ff  er. Sections were incu-bated in this solution for 24 hours at 4 ◦ C, and washedtwice with washing bu ff  er. 2.1.3. Visualization and image acquisition Immunofluorescently labeled sections were dehy-drated in a graded ethanol series, transferred to methylsalicylate, and wet mounted for visualization using aLeica TCS-NT Laser Scanning Confocal Microscope(LSCM) at the W. M. Keck Bioimaging Facility (Keck Lab) at the Arizona State University using Ar andHe  /  Ne lasers. For each sample, 32  z -plane optical sec-tionswereimagedevery0.00125mm(1.25  µ m)througha 0.040mm (40  µ m) region with 0.16mm × 0.16mm(160  µ m × 160  µ m) in the  x –  y  dimensions (see Fig. 1).Serial sections of these images were obtained using C󐁯󐁮􀁦󐁯󐁣󐁡󐁬D󐁡󐁴󐁡S󐁥󐁴󐁯􀁦T󐁩󐁳󐁳􀁵󐁥B󐁩󐁯󐁰󐁳󐁩󐁥󐁳 1󐁭󐁭󰁸1󐁭󐁭󰁸1󐁭󐁭 󰁴󰁩󰁳󰁳󰁵󰁥󲀜􀁢􀁩󰁯󰁰􀁳󐁹󲀝 1󐁭󐁭󰁸1󐁭󐁭󰁸0.06󐁭󐁭 󰁴󰁩󰁳󰁳󰁵󰁥󰁳󰁥󰁣󰁴󰁩􀁯󰁮 󰁃􀁲󰁹 􀁯󰁳󰁴󰁡󰁴󰁣󰁵󰁴󰁳󰁥󰁣󰁴󰁩􀁯󰁮 󰁃 􀁯󰁮󰁦􀁯󰁣󰁡󰁬􀁯􀁰󰁴󰁩󰁣󰁡󰁬󲀜􀁳􀁣􀁡􀁮󲀝 0.00125󐁭󐁭 󰁢󰁥󰁴󰁷󰁥󰁥󰁮󰁩󐁭󰁡􀁧󰁥󰁳 32 􀁯􀁰󰁴󰁩󰁣󰁡󰁬󲀜􀁬􀁡󐁹􀁥󰁲􀁩󰁭􀁡󰁧􀁥􀁳󲀝 0.16󐁭󐁭:󰁸 󰀰 󰀮󰀱 󰀶 􀁭 􀁭 󰀺 􀁹 󰁓 󰁩󰁺󰁥􀁯󰁦󰁥󰁡󰁣󰁨󰁩󐁭󰁡􀁧󰁥󰁩󰁳 1024󰁸1024 􀁰󰁩󰁸 󰁥󰁬󰁳 .U 􀁰󰁴􀁯 12 󰁳󰁣󰁡󰁮󰁳􀁯󰁦󰁥󰁡󰁣󰁨 󰁢󰁩􀁯􀁰󰁳 󰁹 󰁳󰁡󐁭􀁰󰁬󰁥󰁩󰁳 󰁣􀁯󰁬󰁬󰁥󰁣󰁴󰁥󰁤 󰁢󰁹 󰁴󰁨󰁥L󰁡󰁳󰁥 􀁲 󰁓 󰁣󰁡󰁮󰁮󰁩󰁮􀁧 󰁃 􀁯󰁮󰁦􀁯󰁣󰁡󰁬M󰁩󰁣 􀁲 􀁯󰁳󰁣􀁯􀁰󰁥(L 󰁓󰁃 M).󰁔 󰁨󰁥󰁦󰁩󰁮󰁡󰁬󰁩󐁭 󰁡􀁧󰁥󰁩󰁳󰁡󰁬󰁩󰁮󰁥󰁡 􀁲 󰁳󰁴󰁡󰁣󰁫􀁯󰁦32 󰁬󰁡 󰁹 󰁥 􀁲󰁩󐁭 󰁡􀁧󰁥󰁳󰁦􀁲 􀁯 󐁭 󰁡󰁳󰁩󰁮􀁧󰁬󰁥󰁳󰁣󰁡󰁮 .(󲀜􀁢􀁡􀁳􀁥󰁭􀁥􀁮􀁴󰁭􀁥󰁭􀁢󰁲􀁡􀁮􀁥􀁩󰁭􀁡󰁧􀁥󲀝󰁡󰁮󰁤 󲀜􀁓􀁓NF􀁩󰁭􀁡󰁧􀁥󲀝 ) Figure 1: A diagrammatical illustration at di ff  erent stages of the im-age preparation and analysis. From a single “ biopsy ” sample, a thintissue section is prepared by immunostaining, then the LSCM takesup to 12 “ scans ” of image-sets, each with a 63 ×  magnification. Asingle “ scan ” consists of two sets of 32 “ layer-images ” — one for thebasement membrane and the other for the SSNFs — with a spacingof 0.00125mm between layers. Only one set is illustrated here forsimplicity. These two sets of 32 “ layer-images ” are then distributed toindividual CDM and visual classifiers. Once distributed, each classi-fiers stacks these “ layer-images ” linearly to create a single projected“ basement membrane image ” and a single “ SSNF image ” for furtheranalysis. Leica TCS-NT image software. Images were ob-tained at 63 ×  magnification with an image resolutionof 1024 × 1024 pixels to increase the visibility of smalldiameter fibers to facilitate quantification of their den-sity. 2.1.4. Intraepidermal nerve fiber density quantification For each biopsy, up to 12 scans were observed man-ually, and by up to four di ff  erent observers. The num-ber of fibers embeddedinto or penetrating the basementmembrane was visually assessed by examination of thescans. Mean density for each biopsy were then deter-mined, based on the densities for each series of images,and recordedas a numberof intraepidermalnerve fibersper mm of basement membrane length, yielding a meandensity value for the scans of a single biopsy.Although the conventional method is to report epi-dermal length as the reference for calculating the den-sity value, when  the purpose is to examine only the den-sity of intraepidermal nerve fibers, which are terminat-ing into or penetrating the basement membrane , we usebasement membrane length as the reference.3  2.2. Data and tools2.2.1. Definition and terminology Throughoutouranalysis, the data — in a form of dig-ital images — are undergoingmultipleprocessingsteps.The images at di ff  erent steps, and at di ff  erent levels of grouping, therefore need to have specific names to beidentified. The structure of our data set is diagrammat-ically illustrated in Fig. 1, and is defined as follow: (1)therearea totalof52skinpunchbiopsies,or“ biopsies ”;(2) each  biopsy  has up to 12 LSCM “ scans ” at di ff  er-ent locations along the basement membrane section(s);and (3) a single LSCM  scan  creates a set of 32 “ layer-images ” of basement membrane images and 32 “ layer-images ” of SSNF images. These  layer-images  are thendistributed to, and “stacked” into a single “ basement membrane image ” and a single “ SSNF image ” by in-dividual classifiers for analysis. 2.2.2. Steps of Computerized Detection Method  In this specific study, we use a Computerized Detec-tion Method (CDM), derived from the field of astron-omy, to measure SSNF densities in skin punch biopsiesof our volunteer subjects. In the following sections, wedescribethe steps in the semi-automatedCDM analysis,which include: (1) preparation of digitized images by:(a) convertingthe image format; (b)stacking two sets of  layer-images ; (c)smoothingthestacked basementmem-brane image ; and (d) adding a small Gaussian noise-field to the stacked  SSNF image , and similarly smooth-ing it; (2) determination of the location and length of the basement membrane in a  basement membrane im-age ; (3) detection of SSNF segments in an  SSNF image ;and (4) calculation of the median “ SExtractor detectionrate ” for each biopsy. Upon completion of these steps,the CDM results are compared to the visual classifica-tion results for normalization and proper intercompari-son — by converting from  SExtractor detection rate  toestimated  fiber density  — and for assessment of the re-liability of our method. 2.2.3. Astronomical tools In our CDM, we use analysis tools, such as softwareand computerlanguages, that are commonly used in thefield of astronomy. These tools have been developedand optimized over time for image analysis, especiallyfor detecting stars and small distant galaxies in a largefield of view, which somewhat resemble the appearanceof SSNF segments in a basement membrane. Since themajority of astronomical tools are developed primarilyunder the UNIX  /  Linux operating systems, all processesfor this study are performed using Red Hat Linux  R  orthe freely available Community ENTerprise OperatingSystem R  (CentOS R  ). The astronomical tools used inthis study are distributed with  no  or  at low cost  , andmost of them can also be used with other operating sys-tems, such as Microsoft Windows R  or Macintosh R  .Other tools, software, and computerlanguages could besubstituted, given the preference of a CDM user.Astronomical image-processing tools have been de-veloped in the last three or four decades by the NationalAstronomical Observatories 1 , and astronomers all overthe world. New and improved astronomical softwarepackages and tools are continuouslyunderdevelopmentas well. These image-processing tools currently consistof many large software packages, each with 2–4 mil-lion lines of code, and many hundreds of person-yearsof software development. The main tools and softwareused in this project are, amongst others, the Smithso-nian Astrophysical Observatory Image DS9 (SAOIm-age DS9: Joye and Mandel, 2003), the Flexible Im-age Transport System (FITS: Wells et al., 1981), theNational Optical Astronomy Observatory (NOAO) Im-age ReductionandAnalysis Facility (IRAF: Stefl, 1990,and references therein), the Interactive Data Language(IDL R  ), the Source Extractor (SExtractor: Bertin andArnouts, 1996), SuperMongo (SM 2 ), and FORTRAN77 (F77). More detailed descriptions of these tools andsoftware packages are given in the appendices. Most of these software packages and languages are distributedon the internet with low or no cost. Since these toolscould be substituted by many other free software pack-ages and languages,all of the CDM steps canbe in prin-ciple constructed  without any major software cost to theuser  . In this paper, we present one well tested imple-mentation of this CDM. 2.3. Post-digitized image preparation In this section, we give a high level review of all thenecessary steps of the CDM. The specific details, nec-essary to reproduce these steps, are given in the Appen-dices. Readers not interested in the high level steps maydirectly proceed to  § 2.4, or even to  § 3. 1 National Optical Astronomy Observatory (NOAO): http://www.noao.edu/ National Radio Astronomy Observatory (NRAO): http://www.nrao.edu/ Space Telescope Science Institute (STScI): http://www.stsci.edu/ 2 The SM Reference Manual. Robert H. Lupton and Patricia Mon-ger (1977): http://www.astro.princeton.edu/ ∼ rhl/sm/sm.htmlhttp://www.supermongo.net/ 4  2.3.1. Image format conversion The first step in the image preparation is convertingtheimagefileformatasitarrivesformtheLSCM.Whilethe srcinal layer-images obtained with the LSCM aresaved in the Tagged Image File Format (TIFF), theimage-displaying software, DS9, and  all  subsequentsoftwarerequirestheFITSimageformat. Toconvertthefile format, a pre-written F77 program with a sequenceof various Linux commands and IRAF tasks is used.Once an srcinal TIFF image is read in by this pro-gram, the signal (srcinalimage data) is recordedin oneof the Red-Green-Blue (RGB) channels of the portableanymap (PNM) format image, which are then separatedinto three independent PNM image files. Using IRAFtasks, the image file containing the data is saved intoa FITS image. As a result of this file format conver-sion, srcinally green- and red-colored TIFF images forthe basement membrane and SSNFs are converted intogray-scale images as shown in Figs. 2–4. During thisformat conversion, the srcinal signals are rounded intointegervalues in units of Data Numbers (DNs), with thelowest value set to zero. To avoid a systematic errordue to this zero-signal, a value of 1.0DN is added to allpixels. Since the immunostained basement membraneand the SSNFs have large signal values (  1000DN),adding 1.0DN does not a ff  ect the final results. While it would take a long time if all of these pro-cesses are performed manually, this pre-programed im-age conversion only takes a fraction of a second for animage on modern computers.2.3.2. Stacking layer-images and smoothing the base-ment membrane image The SSNFs are running through a basement mem-brane and surrounding tissue not two-dimensionally,but three-dimensionally. A single layer-image thereforecannot capture an entire SSNF, unless it happens to ex-actlylineupwiththescanningfocalplaneoftheLSCM.By imaging at 32 di ff  erent depths of a biopsy sectionfor a single scan, and by stacking these layer-images,we ensure that SSNFs are imaged. Stacking these 32layer-images also helps the CDM detection software byenhancing the total signal-to-noise (S  /  N) level of thebasement membrane and of SSNFs. The second stepin the preparation therefore includes: (1) stacking allthe basement membrane and SSNF layer-images froma single scan; and (2) smoothing the stacked basementmembrane image (see Fig. 2). The stacked SSNF im-age requires several more steps in this preparation, andhence they are treated separately in the following steps( § 2.3.3).The first process is to stack all of the layer-imagesinto a single image using an F77 program. Once a com-puter disk path to a single scan is provided, this pro-gramcreatesanewsub-folderandthenecessaryfiles forsubsequent analysis. The number of layer-images to becombinedis set to 32 for our CDM, but this numbercanbe changed with a few easy modifications to the pro-gram if necessary. To actually stack the layer-images,theIRAF executablefilecreatedbytheF77programhasto be run in the directory where the IRAF parameterfileis stored (usually at the top working directory). Oncethis script has run, IRAF automatically starts up andstacks the 32 layer-images to create a single basementmembraneimage and a single SSNF image. During thisprocess, 3 σ -clipping— where σ stands forthe standarddeviationof the imagenoise — is appliedto removeanyhot or bad pixels. These pixels usually have extremelyhigh or low values due to spontaneous electrical e ff  ectsof a photodetector, e.g., photomultiplier tubes (PMTs)orCharge-CoupledDevices(CCDs), usedintheLSCM,and would cause a systematic error in the subsequentanalysis.The next process is to outline the basement mem-brane in the stacked image. Fig. 2a shows the stackedbasement membrane image without any treatment (i.e.,smoothing),with green-coloredcontoursdrawnat a sig-nallevelof130.0DN.Withoutsmoothing,theperimetercontours outlining the basement membranehave jaggededges, which will cause errors in subsequent SSNF de-tection at the edges of the basement membrane. Sincethe inner and outer edges of the basement membraneshould be ideally outlined by rather simple contours,we smooth the stacked basement membrane image toremove this graininess. Fig. 2b shows the smoothed im-age of Fig. 2a, with the green-colored contours appliedto the same signal level of 130.0DN. While the srci-nal image (Fig. 2a) has a slightly sharper appearance,the edges in the smoothed image (Fig. 2b) are mucheasier to trace and visualize. Another important e ff  ectof smoothing is that it softens and removes, or shrink the regions of slightly lower signals within a basementmembrane. Without any treatment, these low-signal re-gions appear as small “bubbles” along edges or insideof basement membrane (see Fig. 2a), and would a ff  ectthe final results. In the same IRAF script to stack layer-images, another IRAF task —  BOXCAR  with a box-sizeof 7 pixels — is therefore used to smooth the stackedbasement membrane image.  As in the previousstep, stacking multiple images, cre-ating folders and files, and smoothing the stacked base-ment membrane image only take at most a few secondson a  ∼ 2GHz Linux box, while the same processes per- 5
Similar documents
View more...
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks