3D DECONVOLUTION IN DYNAMIC MICROSCOPY
Contents
1. EIDGENOSSISCHE TECHNISCHE HOCHSCHULE LAUSANNE POLITECNICO FEDERALE LOSANNA SWISS FEDERAL INSTITUTE OF TECHNOLOGY ECOLE POLYTECHNIQUE SECTION MICROTECHNIQUE FEDERALE DE LAUSAN NE 3D DECONVOLUTION IN DYNAMIC MICROSCOPY USER MANUAL Pierre Besson Supervisor C dric Vonesch Professor M Unser I INSTALLATION ImageJ installation and the plugin installation procedure are described on http bigwww epfl ch demo deconvolution3D Il PSF GENERATOR The plugins PSF Simple and PSF Gibson are used for simulating the behavior of a transmitted brightfield optical microscope They may be launched from the plugin menu A PSF Simple Model A PSF Simple Generator v1 0 OPTICS PARAMETERS NA 1 4 n 1 515 The simplified model does not df 800 um take into account the non stationarity of the PSF in the space Thus generated PSF has the focal plane as symmetry plane and optical axis as y 128 symmetry axis This model is to be used when WINDOW PARAMETERS Image Size in pixels x 128 Slice 16 the knowledge about the um pixel in object plane parameters is not precise and 1 0 uses only 4 parameters that are generally known y 1 0 z step um slice 0 5 Light wavelength 450 nm Close Credits Process B PSF Gibson s Model PSF Gibson Generator v1 0 OPTICS PARAMETERS NA 1 4 ti 0 19 mm ni 1518 ni 1518 tg 0 17 mm tg 0 17 mm ng2. to deal with much larger stacks 1024x1024x102 32 bits gray scale Operations may be very time consuming For example with an input stack 1024x1024x20 32 bits gray scale 80 MB Algorithm Time Constraint Inverse 55 min I i i terative deconvolution 40 Lh 45 min iterations ForWaRD 4 h 30 min IV ODELETTES PLUGIN A Screenshot Wavelets based Deconvolution PSF blabla Algorithm ForWaRD E Scale reduction 1 Epsilon 1 10E 9 Credits Close Convolve Deconvolve B How to use The principle was described in section III B Scale reduction is the number of wavelet transform applied on the stack of image NOTE For all the dimensions in pixels make sure that dim scale reduction 1 is still integer For example input stack 10 10 16 The highest Scale reduction that can be used is 1 Others notes are the same
3. Lais ng Lais O ns 1 46 ts 20 21 pm zs 0 2 um M WINDOW PARAMETERS Image Size in pixels x 128 y 128 Slice 16 um pixel in object plane x 1 0 y 1 0 z step um slice 0 5 Light wavelength 450 nm Close Credits Process Contrary to the simplified model Gibson s model takes into account the non stationarity of the PSF The focal plane may not be a symmetry plane anymore Hence the PSF should be chosen in order to give optimum results at the middle of the stack Parameters Optics parameters Model Ge SE Simplified Gibson NA Numerical x x aperture Refractive index n of immersion X medium df Focal length of x objective lens ti See schemes X ni lt X ni a X tg S X tg S A ng X ng S A ns lt X ts S A ZS id X M Magnification X Label Name in the PSF plugin Detail Optical meaning of the label Model A cross shows that this parameter is used for the model Scheme of Gibson s parameters Objective lens Immersion medium ni Ontical axis 7 Parameters Meaning ti Width of immersion index ni Refractive index of immersion index tg Width of cover slip ng Refractive index of cover slip na Depth of the region of interest in the specimen Depth to which starts the recording e may be negative Index with are the designed perfect parameters while
4. the desired action mA BWN ra NOTE Epsilon has to be small and is used to perform a stable division in the Fourier Space Default value is recommended Lambda is used for Constraint Inverse Deconvolution For noise free signal lambda should be small 0 1 For a very noisy signal lambda should be large 0 8 Gamma is used to ensure the convergence of Van Cittert s algorithm Using a PSF generated above the condition is Gamma lt 1 slice If you use a measured PSF you should enter a lower Gamma may be 0 8 Slice Iteration is the number of iterations proceeds during Van Cittert s algorithm Larger is slice larger should be Iteration in order to obtain good results Click Refresh to check new images and actualize the list THE X AND Y DIMENSIONS OF THE STACK TO BE DECONVOLVED MUST BE DYADIC ELSE THE ALGORITHMS DO NOT WORK THIS IS DUE TO THE FAST FOURIER TRANSFORM NOTE THAT IF THE NUMBER OF SLICE IS ALSO A DYADIC NUMBER THEN THE ALGORITHMS ARE MUCH FASTER C Memory and time Depending on the size of the stack of image the time and the memory needed for the process can be large To be able to run the plugin on large images more memory should be given to ImageJ Please set the parameter Maximum Memory Edit gt Options gt Memory to at least 640 MB With this setting you are able to deal with images up to 64MB 512 x 512 pixels 64 slices 32 bits Setting memory at 1 3 GB allows us
5. with no star means actual Often it is considered that actual parameters are perfect Window parameters Parameters Meaning a D Dimensions x and y of the PSF in Image Size in pixels y GE Z Number of slices D Size of a CCD sensor in the ym pixel in y object plane real size M Object Plane Step between two consecutive images Light wavelength Wavelength of the emitted light NOTE The focal plane or its closest plane is at the top of the stack The remainder is constructed cyclically Itis recommended to save the created PSF Creating a PSF may be time consuming Conditions on parameters Simplified model o xand y sizes of the image in pixels have to be dyadic numbers 2 Gibson s model o xand y sizes of the image in pixels have to be even numbers C How to use Choose a model then type your own parameters Press Process to create the PSF Close to quit or Credits to see the credits UL DECONNOLUTION PLUGIN A Screenshot oc Deconvolution PSF blabla J Refresh Algorithm Inverse Filtering HH Epsilon G 10E 9 Lambda 075 Gamma 0 5 Iteration 25 Credits Close Convolve Deconvolve B How to use Once you have created your PSF Open the image to deconvolve or convolve Ensure that the generated PSF is selected as the PSF Choose an algorithm Enter the corresponding constant Press
Download Pdf Manuals
Related Search