Grid Assembly. User guide. A plugin developed for microscopy non-overlapping images stitching, for the public-domain image analysis package ImageJ

Transcription

1 BIOIMAGING AND OPTIC PLATFORM Grid Assembly A plugin developed for microscopy non-overlapping images stitching, for the public-domain image analysis package ImageJ User guide March 2008

2 Introduction In microscopy one has often to find a compromise between high resolution and a large field of view. For applications that need both high resolution and large field of view, a possibility to overcome this limitation is to record the object in multiple tiles and then to stitch them together in the computer. Grid_Assembly was developed to answer to the exigency of obtaining single mosaics of large dimension, from the union of sets of non-overlapping tiles acquired in grid-mode. The resulting mosaic is typically subject to some artefacts, as seam effects and uneven brightness, which are essentially linked to the acquisition-system side. These effects are principally due to the uneven illumination of the single tiles during the image acquisition and to some inclination of the whole sample respect to the objective. These factors can cause brightness variations at a local level (single tiles) as well as at a global level (complete mosaic). Brightness discontinuities between contiguous tiles originate the typical seam effects. Therefore, some post-processing of the images seems to be helpful and necessary in order to obtain agreeable mosaics and in prevision of a following mosaic deconvolution. In fact the deconvolution exalts the high-frequencies component, thus making worse the seam artefacts. Grid_Assembly stitches together the input images, by using the scanning stage-positions information contained in the images names. It can perform background correction by two different methods, one of which is theoretically compatible with image quantification operations. It can perform a local seam correction. If both the background correction and the seam correction are selected, the seam correction is performed on the mosaic corrected for the background.

3 Index Insatallation... 1 Plugin structure... 1 Input/Output... 2 Input... 2 Output... 2 Name convention and scanning path... 4 Name convention... 4 Scanning path... 4 Post-processing... 5 Background correction fit parabola... 5 Background correction filtering... 7 Seamless correction... 8 Assembly examples Deconvolution... 13

4 Installation Download the jar archive Grid_Assembly_.jar and save it in the plugins folder of ImageJ. Start ImageJ. Under the menu Plugins you will find the voice Grid Assembly. To start the plugin, click on Grid Assembly > Assembly. Plugin structure The plugin Grid Assembly allows you to stitch together non-overlapping microscopy images acquired in grid mode. From the single tiles, a single large mosaic is created. When you launch the plugin, a user-friendly interface appears. The interface is structured in two menus: the first, Assembly, allows you to set the input and the output folder, the grid-scanning modality, the dimensions of the final mosaic, and to define the convention for the images names. Actually Grid Assembly gets the information about the scanning stage position, the z-plane and the channel from the name of the files in input. The last panel of the menu Assembly allows you to eventually ask for background correction and/or seam correction. The background correction compensates for the uneven illumination of the samples, the seam correction reduces the abrupt transitions (seams) between contiguous tiles in the resulting mosaic. When something different from None is selected in the pull-down menus Background corr. or Seamless corr., the general menu Post-processing settings changes colour and becomes accessible. The menu Post-processing settings allows a finer tuning of the parameters for the background correction and the seam correction. Press the button Assembly to start the processing; press the button Close to close the interface. Note that when you launch the Grid_Assembly plugin, the settings of the previous job are loaded. The settings are memorized in the file tempassembly.cfg, which is automatically generated in your ImageJ folder the first time that you run the plugin, so please do not delete it. 1

5 Input/Output Input The Directories panel in the Assembly menu allows you to select the directory with the input images (source) and the directory where the results will be saved (destination). The input folder must contain only the images to be processed. All the images are supposed to have the same name structure and to contain at least the number of the stage position. If more than one Z-plane and/or more than one channel are present, it is supposed that the images names contain also the number of the Z-plane and/or of the channel. The order numbers that indicate the stage position, the Z-plane and the channel are supposed to be sequential. Only two-dimensional images are supported. The application handles volumetric and multi-channels acquisitions, but the stacks have to be converted to individual images before running the plugin. All the 2Dimage formats supported by ImageJ are supported. If more than one channel is present, the different channels are processed independently. Remember that the available memory for ImageJ is typically 2/3 of the physical memory of your computer, so you could get Out-Of-Memory exception if you try to stitch heavy datasets. Moreover, if some post-processing is performed, the memory consumption increases significantly. To increase the available memory, select ImageJ Edit menu > Options > Memory & Threads and set a new value for Maximum Memory (always inferior to the physical memory of your computer), then restart ImageJ. 2

6 Output Assembly outputs one tiff image, 32 bit, for each channel. If more than one plane exists in the Z dimension, the output is one 32 bit tiff stack for each channel. The resulting mosaic relative to the first Z-plane, first channel, is displayed at the end of the processing. All the results are saved in the destination folder with the name assembly-ch. 3

7 Name convention and scanning path Name convention It is supposed that the information about the position of the single tile in the global three-dimensional acquisition grid is contained in its name. The panel Name convention in the Assembly menu allows you to define the correspondence between the different parts of the image name and the information about the position of the tile in the plane (its positional order number), its position in the third dimension (Z) and its channel. When the source directory is selected, the name of the first image in the folder appears in the dedicated fields. On the name that appears in the second field, you have to select with the mouse the numbers that correspond to the position, the Z and the channel and to press the correspondent button to memorize the localization of the information in the name. When the Define Position, Define Z or Define Channel is pressed, the selected numbers are highlighted in the corresponding colour in the first field (name in bold letters). The first field allows you to check your selection. If only one Z-plane and/or one channel are present, simply do not select anything for the Z and/or the Channel. The Clear button allows you to clear your selections. Scanning path In the Scanning path panel you have to define the numbers of tiles in the X direction (Columns) and in the Y direction (Rows) of the final mosaic. Moreover, press one of the gridlike scanning path button (the plugin supports horizontal or vertical, parallel or anti-parallel, normal or flipped acquisition scheme). 4

8 Post-processing settings In the Post-processing settings panel you can set the parameters of the correction algorithms, that is the background correction and/or the seam correction algorithms. Two techniques are implemented for the background correction, the parabola-fitting and the filtering: you can choose one of the two. The technique for the seam correction is unique. Background correction fit parabola This method relies on the hypothesis that the uneven illumination of the single tiles could be modelled through a paraboloid, which can usually be accepted only in bright field microscopy. Relying on this hypothesis, the correction algorithm evaluates on the tile the parabola that can better fit the non-uniformity of the illumination and subtracts the so-estimated background to the original tile. If one can trust on the physical hypothesis concerning the background (parabolic model), quantification operations can be performed on the resulting image even if different values have been subtracted from the different pixels of the image. It is important to underline that the evaluation of the background can fail if details of relative big dimension are present in the image, or if part of the image is constituted by black background. In fact the parabola estimation can be significantly affected by the presence of big details compared to the dimension of the tile or by significant black areas, thus giving back a false estimation of the real background. Examples 1 and 2 show an unreliable and a good estimation of the background from a tile. In the Background correction: Fit parabola panel of the Post-processing menu you can choose between different ways to evaluate the background. The evaluation of the parabola can be performed 5

9 tile-by-tile (option Local, which is the default option) or on an image selected by the user (option From selected image ), typically a blank image acquired on purpose for the background estimation. In this last case you have to select the specific image for the background evaluation too. Selecting the option Auto, the correction algorithm looks automatically for the best estimation of the background between the parabola evaluations on all the single tiles of the mosaic (first Z-plane and first channel). If the Auto or the From selected image option is selected, the same background will be subtracted to all the tiles of the mosaic: this operation relies on the hypothesis that the uneven illumination of the investigated area is exactly the same for all the tiles. In case of subtraction of the same background from all the tiles, it has to be considered that the subtraction will probably create some artefacts in eventual uniform black areas which are not affected by uneven illumination artefact. For example, in the following image the subtraction of the estimated background creates an artificial light area in the right upper corner of the image. Concerning the automatic evaluation of the best background, if there are many images in the mosaic with large details or large black areas, which are elements that badly affect the real background estimation given by the uneven illumination, the estimation of the best background will be deviated. The parabola subtraction is performed tile-by-tile and does not correct for any global trend of brightness variation across the whole mosaic, bust just for the local uneven illumination (single tile). Example 1: the black area does not allow a correct evaluation of the uneven illumination of the sample. Example 2: reliable estimation of the background 6

10 Background correction filtering This method relies on the hypothesis that the single tiles backgrounds and the brightness changes in the global mosaic are smooth and slowly-varying. The correction algorithm removes from the assembled mosaic a band of low frequencies whose position and extension depend on the tiles size and can be tuned by the user. However, no specific frequencies values are asked, because the filtering is performed in the spatial domain through an opportune combination of Gaussian filtering and simple images arithmetic operations. The parameters of the Gaussian filtering (the kernel opening σ) are linked to the dimension of the tiles and some default parameters are given (to set the default values, press the Set Default button). You can act on the strength of the correction, that is to say on the position and on the extension of the removed frequencies band, setting the two parameters Correction strength [%] and Lower frequencies Ratio. The parameter Correction strength [%] has to be set as a percentage of the tile dimension and is linked to the position of the removed frequencies band. The default value is 88% of the tile size and it can be thought as the spatial semi-period of variation of the background function. This parameter can be increased if a stronger correction is needed, but the more it is increased, the more real low-frequency components of the image may be lost. The second parameter, the Lower frequencies Ratio, must be bigger than 1 and is linked to the extension of the removed frequency band. Its default value is 2. This parameter is linked to the distance between the two bounds of the subtracted frequency band and you can set a higher value if you want to remove a larger band of frequencies. The frequency correction algorithm corrects for the local and the global uneven brightness. It should not be used if some quantification has to be done on the image. The example 3 shows a non-corrected mosaic, the low frequencies image that is subtracted from the original one and the mosaic corrected for the background through the filtering technique. 7

11 Example 3: background correction, filtering technique Seamless correction The seam artefact, that is the presence of abrupt and evident transitions between contiguous tiles, is due to the non-homogeneity of the illumination on the single tiles as well as on the global mosaic area. Consequently, this artefact is partially attenuated after the background correction. The seam correction algorithm performs a further smoothing of the transitions between the tiles. In the assembled mosaic, the seams are high frequency components: the principle of the seam correction algorithm is thus to filter these frequency components, without affecting the real high-frequency content of the image. The correction is performed in the spatial domain, through an opportune combination of Gaussian filtering and simple images arithmetic operations, so that no exact frequency 8

12 values have to be set. The parameter for the Gaussian filtering is linked to the tile dimension and a default value is given ( Correction strength [%] = 3). The parameter Correction strength [%] is set as percentage of the tile dimension and it is linked to the period of the spatial variations in the mosaic that will be corrected. It can be increased, but too high values of this parameter could imply a loss of definition of the high frequency details of the mosaic, close to the seams. If the Correction strength [%] is kept low, the seam correction algorithm does not affect the centres of the tiles, but just their borders, so it can be applied also if one want to perform quantification operations on the mosaic. Example 4 shows the result of the seam correction in a detail of a mosaic. Example 4: seam correction 9

13 Assembly examples Some examples are presented to help you in the choice of the better correction setting. Example A local parabola background correction Original mosaic and corrected mosaic through background correction (parabola fitting, local parabola estimation) and seam correction. The correction for the background by subtraction of the locally estimated parabola fails in the parts where large black areas are present. The seam correction properly smooths the transitions between contiguous tiles. 10

14 Example B seam correction Example C background correction techniques Non- corrected mosaic. 11

15 Background correction by best parabola subtraction (automatically estimated) and seam correction (default parameters) have been applied to the original mosaic. Frequency background correction and seam correction (default parameters) have been applied to the original mosaic. 12

16 Deconvolution Concerning the deconvolution, we suggest to stitch the tiles together before, the to perform the deconvolution. An example is reported: Deconvolution of the single tiles (Huygens, CLME algorithm, in/near object background estimation method), then assemblage of the mosaic. Deconvolution of the assembled mosaic (Huygens, CLME algorithm, in/near object background estimation method). 13

17 Deconvolution of the single tiles (Huygens, CLME algorithm, in/near object background estimation method), then assemblage of the mosaic: detail. Deconvolution of the assembled mosaic (Huygens, CLME algorithm, in/near object background estimation method): detail. 14