Cell uptake 2 channels coloc
Uptake assay with two spot channels and brightfield cell.
More details
Uptake assay with two spot channels and brightfield cell.
Contributed by: Melanie Schuerz (University of Salzburg)
Downloads
Ready-made project templates and AI models you can drop straight into EVAnalyzer.
Project Templates
Section titled “Project Templates”Full .evatemplate project templates - classes, plate settings and pipelines - ready to open with EVA › Open…, EVA › New from template… or New project from a template… on the Start page. Templates with example images come as a .zip too. See Saving a Project as a Template for how these work.
Cell uptake 3 channels coloc
Uptake assay with two spot channels and brightfield cell.
More details
Uptake assay with two spot channels and brightfield cell.
Contributed by: Melanie Schuerz (University of Salzburg)
Coloc 2 channels
Spot colocalization with two image channels.
More details
Spot colocalization with two image channels.
Contributed by: Melanie Schuerz (University of Salzburg)
Coloc 3 channels
Spot colocalization with three image channels.
More details
Spot colocalization with three image channels.
Contributed by: Melanie Schuerz (University of Salzburg)
EV uptake in cells
Methods of Cell Biology course
More details
Methods of Cell Biology course
Contributed by: Melanie Schuerz (University of Salzburg)
Example Colocalization
Ported from CellProfiler ExampleColocalization pipeline
More details
Two-channel spot colocalization (Cy3-like / Cy5-like nucleosome staining). Ported from the CellProfiler ExampleColocalization.cppipe example. Channel 0 = OrigStain1 (files *_N_R), Channel 1 = OrigStain2 (files *_N_G), merged into one 2-channel TIFF per site via Fiji. Object counts run ~10-15% off CP and coloc% ~5pt high, mainly because evanalyzer’s watershed declumping (prominence-based) splits touching spots differently than CP’s fixed-radius local-maxima method, and evanalyzer’s illumination correction averages over blocks before fitting (more robust to bright spots) instead of fitting the raw pixels directly like CP does.
Contributed by: Jeff Reifenberger (Massachusetts General Hospital), Brad Berstein’s (Massachusetts General Hospital)
About the example data
Measuring the colocalization between fluorescently labeled molecules is a widely used approach to measure the degree of spatial coincidence and potential interactions among subcellular species (e.g., proteins). This example shows how the object identifcation and RelateObjects modules are used to measure the degree of overlap between two fluorescent channels.
About these images:
Fluoresecent images of a histone-modified nucleosome that is labeled with a Cy3-like dye and an antibody labeled with a Cy5-like dye that is sensitive to to the histone modifications.
These images were contributed by Jeff Reifenberger of Brad Berstein’s Lab at Massachusetts General Hospital, March 2012
Example Comet Assay
Ported from CellProfiler ExampleCometAssay pipeline
More details
Single-channel comet assay (DNA damage electrophoresis). Ported from the CellProfiler ExampleCometAssay.cppipe example. Comet = whole cell blob (head+tail), CometHead = dense head sub-region, CometTail = Comet minus CometHead. CP masks the image to Comet before re-thresholding for CometHead (MaskImage) and has no direct evanalyzer equivalent for image-level masking by object class; approximated here with an intersection filter (classifyObjects overlappingWith comet@whole) instead, applied after thresholding the full image. CP’s ‘Robust Background’ threshold method also has no evanalyzer equivalent; approximated with Otsu. Illumination correction’s median-smoothing radius needed a much larger value (20 block-grid units) than a naive unit conversion from CP’s settings suggested, otherwise residual block-boundary noise fragmented each comet into many tiny pieces instead of one blob. Result: CometTails.tif matches CP exactly (6/6/6 whole/head/tail); NoTails.tif is close but not exact (5/5/5 vs CP’s 6/4/6) - one faint comet isn’t detected, and because head-detection here runs on the full image rather than a CP-style pre-masked one, all 5 found comets get a head where CP only assigned heads to 4 of its 6.
Contributed by: Scott Floyd (Centro de Protección e Higiene de Las Radiaciones), Michael Pacold (Centro de Protección e Higiene de Las Radiaciones), Jorge Ernesto González (Centro de Protección e Higiene de Las Radiaciones)
About the example data
The fluorescent comet images were donated by Scott Floyd and Michael Pacold.
The silver-stained comets were contributed by Jorge Ernesto González from the Centro de Protección e Higiene de Las Radiaciones (CPHR).
Example Illumination Correction 1
Ported from CellProfiler ExampleIlluminationCorrection_Example1_EachMethod pipeline
More details
Illumination correction of one Transfluor GFP image: Regular method, median smoothing, divide. Ported from CellProfiler ExampleIlluminationCorrection_Example1_EachMethod.cppipe. CP’s median filter size 40 is a disk of radius 17 px at full resolution; here block size 1 + median radius 17 (square window). The corrected image (Save Image) correlates r=0.99 with CP’s. Example1_AllMethod (one function averaged over all 72 images) is not ported: evanalyzer calculates the correction per image.
About the example data
Illumination correction is often important for both accurate segmentation and for intensity measurements. This example shows how the CorrectIlluminationCalculate and CorrectIlluminationApply modules are used to compensate for the non-uniformities in illumination often present in microscopy images.
This part corrects each image with its own illumination function (Regular method, median smoothing, divide).
About these images:
The GFP channel from the SBS Roche Transfluor image set, available from http://www.broadinstitute.org/bbbc/sbs_roche_transfluor.html.
Example Illumination Correction 2
Ported from CellProfiler ExampleIlluminationCorrection_Example2 pipeline
More details
Nuclei (mCherry) found three times: on the raw image, after a too-small background correction (block 8) and after a suitable one (block 20, then Li/Minimum Cross-Entropy threshold). Ported from CellProfiler ExampleIlluminationCorrection_Example2.cppipe. CP’s threshold is calculated on the unsmoothed image and applied to a sigma-1 blurred copy (Image Cache + value source). Its ‘Intensity’ declumping is Watershed with intensity seeds, 14 px maxima suppression (CP’s low-res value); seed smoothing sigma 2.85 instead of CP’s 5.7 was chosen empirically, it splits touching nuclei closer to CP. evanalyzer’s Subtract keeps the image mean while CP subtracts the whole background and clips at 0, and its median smoothing works on the block grid (radius 1 block) instead of CP’s full-resolution disk. Result vs CP: uncorrected 40 vs 42 objects, small block 9 vs 21 (CP’s extra objects are fragments of 2-53 px), large block 24 vs 30 (objects > 50 px: 23 vs 26, mean area 839 vs 844 px). The example’s point - only the large block correction yields proper nuclei - is reproduced.
About the example data
Illumination correction is often important for both accurate segmentation and for intensity measurements. This example shows how the CorrectIlluminationCalculate and CorrectIlluminationApply modules are used to compensate for the non-uniformities in illumination often present in microscopy images.
This part finds nuclei three times - on the raw image, after a too small and after a suitable background correction - to show how the correction affects segmentation.
About these images:
The image of the CellProfiler example ExampleIlluminationCorrection prefixed with “—W00001—”.
Example Illumination Correction 3
Ported from CellProfiler ExampleIlluminationCorrection_Example3 pipeline
More details
Worms in a well (bright-field, dark worms). Ported from CellProfiler ExampleIlluminationCorrection_Example3.cppipe. The well is found with 3-class Otsu (middle class foreground) and shrunk by 5 px; the inverted image is background-corrected with block 2 + polynomial fit and subtracted. Well area 172121 vs CP 172394 px, shrunken well 164654 vs 163852 px. CP restricts the polynomial fit to the shrunken well (MaskImage); evanalyzer cannot mask an image by objects, so the fit uses the whole image - the corrected image still correlates r=0.96 with CP’s inside the well. Not ported: the second correction (Regular + Convex Hull smoothing + divide), evanalyzer has no convex-hull smoothing.
About the example data
Illumination correction is often important for both accurate segmentation and for intensity measurements. This example shows how the CorrectIlluminationCalculate and CorrectIlluminationApply modules are used to compensate for the non-uniformities in illumination often present in microscopy images.
This part finds the well and corrects the uneven background of a bright-field image of worms inside it.
About these images:
The image of the CellProfiler example ExampleIlluminationCorrection prefixed with “ADSAStaphInfection2_”.
Example Percent Positive
Ported from CellProfiler ExamplePercentPositive pipeline
More details
Two-channel percent-positive assay (HT29 cells, DNA + phospho-histone H3 mitosis marker). Ported from the CellProfiler ExamplePercentPositive.cppipe example. Channel 0 = OrigBlue/DNA (files *d0.tif), channel 1 = OrigGreen/PH3 (files d1.tif), merged into one 2-channel TIFF per site via Fiji (images/). Mapping: IdentifyPrimaryObjects -> Gaussian blur (sigma 1.0 = CP threshold smoothing scale 1.3488/0.6744/2) + Threshold computed on the raw image (LI = Minimum Cross-Entropy for nuclei; 3-class Otsu, middle class foreground, lower bound 0.05 for PH3) + connected components + watershed (Shape -> distance-map seeds, Intensity -> intensity seeds with CP’s maxima-suppression radius min-diameter/1.5 and the effective sigma 2.83 of CP’s truncated 10-px smoothing kernel) + fill object holes (CP: after declumping only) + area filter from the typical diameter (pid^2/4) with border objects discarded. RelateObjects + FilterObjects(Children_PH3_Count >= 1) -> class ph3_pos_nuclei (nuclei overlapping a ph3 object). ClassifyObjects(MaxIntensity_OrigGreen > 0.2) -> class ph3_pos_by_intensity. CalculateMath (percent positive) has no pipeline equivalent: compute it after export as count(ph3_pos_nuclei) / count(nuclei) * 100. Result vs CP 4.2.8: nuclei 281 vs 280 (median area 115 vs 118.5 px), PH3 19 vs 20 (median 91 vs 94.5 px), PH3-positive nuclei 20 vs 20, percent positive 7.12 % vs 7.14 %, positive by max intensity 7.47 % vs 7.50 %. Remaining deviations: evanalyzer’s distance-map watershed (prominence-based) splits a few large nucleus clumps that CP keeps whole (largest nucleus 354 vs 471 px), and splits one large PH3 object CP keeps whole (largest 121 vs 191 px); CP’s 3-class Otsu uses a 128-bin histogram.
About the example data
Percentage of nuclei that are positive for the mitotic marker phospho-histone H3 (PH3) in HT29 cells.
About these images:
Image from Jason’s mitotic index experiment (CellProfiler example ExamplePercentPositive). The two CellProfiler input files (*d0.tif = DNA, *d1.tif = PH3) are merged into one 2-channel TIFF: channel 0 = DNA, channel 1 = PH3.
Example Speckles
Ported from CellProfiler ExampleSpeckles pipeline
More details
H2AX foci (speckles) per nucleus, ported from the CellProfiler ExampleSpeckles.cppipe example. Channel 0 = Hoechst (1-162hrhoe2.tif), channel 1 = H2AX (1-162hrh2ax2.tif), merged into one 2-channel TIFF (images/) with Fiji. Nuclei: Gaussian sigma 1 + Otsu (threshold computed on the raw image, like CP), fill holes, distance-map watershed (tolerance 2), area for 120-300 px diameter, edge objects dropped. Foci: CP’s EnhanceOrSuppressFeatures ‘Speckles’ (size 10) is a white top-hat with a radius-5 disk - built here as Image Cache + grayscale Open (11 px ellipse) + Image Math subtract. Then Robust Background threshold, intensity-seeded watershed (min distance 4), and Object Math AND with the nuclei to keep only the part of each focus inside a nucleus (CP’s MaskImage + RelateObjects). Known difference: CP masks the image to the nuclei before thresholding, so its Robust Background statistics only see nuclear pixels; evanalyzer cannot mask an image by objects, so the statistics see the whole image. To compensate, deviations above average was raised from CP’s 2.0 to 4.8 (reproduces CP’s threshold on this image; will need re-checking on other images). Result vs CP: nuclei 6/6, areas within 1%; foci 157 vs 152, median area 56 vs 54.5 px, per-nucleus foci counts within +-2.
About the example data
This pipeline shows how to identify smaller objects (foci) within larger objects (nuclei) and how to relate the two, e.g. to count the foci per nucleus.
About these images:
From the CellProfiler example ExampleSpeckles. The two CellProfiler input files (1-162hrhoe2.tif = Hoechst, 1-162hrh2ax2.tif = H2AX) are merged into one 2-channel TIFF: channel 0 = Hoechst, channel 1 = H2AX.
Example Track Objects
Ported from CellProfiler ExampleTrackObjects pipeline
More details
Drosophila embryo GFP-histone time-lapse: segment embryos per frame and track them. Ported from the CellProfiler ExampleTrackObjects.cppipe example. Input: images/SequenceN.tif, one gray 21-frame time stack per sequence, built with Fiji from the RGB frames as (R+G+B)/3 (= CP ColorToGray ‘Combine’, equal weights); each sequence is its own image, which replaces CP’s per-folder grouping. Embryos = CP IdentifyPrimaryObjects: Gaussian sigma 1 (CP threshold smoothing scale 1.3488), 2-class Otsu computed on the raw image (lower bound 0.01), fill holes, intensity-seeded watershed with CP’s automatic declumping values (seed smoothing sigma 7.14, maxima suppression 16.7 px; CP’s low-resolution speed-up is not reproduced), minimum equivalent diameter 25 px (area >= 491 px), border objects discarded. Tracking: SIMPLE_TRACKER (nearest neighbour, max 50 px) instead of CP’s Overlap method. Result vs CP: objects 63/120/115 vs 65/119/109 per sequence, every CP object matched within 15 px in Seq1/Seq3 and 116/119 in Seq2, median per-object area ratio 0.99-1.00. Tracks: Seq1 3 vs 4 (CP’s extra one is a one-frame object), Seq3 7 vs 7; Seq2 10 vs 6 because CP’s Overlap gives both daughters of a dividing embryo the parent’s label, whereas nearest-neighbour tracking continues the parent in one daughter and starts a new track for the other, plus one frame (t=4) where two touching embryos are merged. Track data: evanalyzer cli export csv --group-by track --class Embryos gives a time course per track (area, shape, intensity per frame); the per-object export has T and Track ID columns. CP’s per-track motion measurements (Displacement, IntegratedDistance, Linearity) are not computed by evanalyzer; they can be derived from the centroids in export parquet and then agree with CP for tracks without a division (e.g. Seq3 displacement 35.8/43.2/51.8/59.1 px CP vs 34.7/42.8/50.5/59.1 px eva).
Contributed by: Victoria Foe, G. Odell, Rob Saint
About the example data
This example identifies the nuclei of a time-lapse movie frame by frame and tracks them from frame to frame. Each sequence is analyzed independently; export the tracks with evanalyzer cli export csv --group-by track.
About these images:
A portion of a time lapse movie of a syncytial blastoderm stage Drosophila embryo with a GFP-histone gene which renders chromatin fluorescent in live embryos. The movie shows nuclear divisions 10 through 13.
Victoria Foe made this movie on a Bio-Rad Radiance 2000 laser scanning confocal microscope using a 40X 1.3NA oil objective. The frames are 7 seconds apart.
GFP-histone transformed files provided by Rob Saint.
V. Foe and G. Odell, 26 July 2001
The 21 RGB frames of each CellProfiler sequence folder are combined into one gray time stack per sequence (Sequence1.tif … Sequence3.tif, gray = (R+G+B)/3 like CellProfiler’s ColorToGray ‘Combine’).
Example Vitra
Ported from CellProfiler ExampleVitra pipeline (nucleus/cytoplasm protein translocation)
More details
Nucleus/cytoplasm protein translocation assay (BBBC014), ported from CellProfiler’s ExampleVitra.cppipe. Input: one 4-channel TIFF per well in images/ (Fiji merge): ch0 = DNA (“Channel 2-”), ch1 = protein (“Channel 1-”), ch2 = 1/VitraChannel2ILLUM, ch3 = 1/VitraChannel1ILLUM. CP divides by precomputed illumination functions (.npy, 1.0-2.0 across the field) which evanalyzer cannot load, so the inverse functions are stored as 8-bit extra channels (round(255/f), <0.4 % error) and applied with Image Math MULTIPLY - same correction as CP, not re-estimated. Nuclei: LI (= CP Minimum Cross-Entropy) threshold computed on the corrected, unsmoothed image (Image Cache + value source) and applied to a sigma-1 Gaussian blur (CP threshold smoothing 1.3488), fill holes, intensity-seeded watershed with CP’s automatic values for 5-30 px diameter (maxima suppression 3.33 px, total smoothing sigma 1.43 = 1.0 + 1.02 in quadrature), area 19.6-706.9 px (diameter range), border objects discarded. Threshold mask agrees with CP on 99.7 % of pixels. Cells (CP Distance-B, 8 px): protein foreground = LI threshold of corrected protein; cell region = (nuclei dilated 8 px AND protein foreground) OR nuclei; Voronoi from the nuclei inside that region. Cytoplasm = cells minus nuclei eroded by 1 px (CP ‘shrink smaller object’). Result vs CP: about 22 % fewer nuclei (577/619 vs 744/782). Evanalyzer’s watershed (seeds from intensity, flood on the distance map with 1 px lines) leaves 13-15 dense clumps per image unsplit, which then fail the 30 px diameter limit; lower seed tolerances raise the count but cut single nuclei into fragments. Matched nuclei have about the same area (median eva/CP 0.94-0.95). Cells/cytoplasm of isolated nuclei match CP; in crowded areas they are larger because the missing clump nuclei leave room for their neighbours. The CalculateMath ratios (nucleus/cell and cytoplasm/cell mean protein, computed after export) match closely: median 1.16 vs 1.15 and 1.01 vs 1.03 (nucleus/cell), 0.83 vs 0.80 and 1.00 vs 1.01 (cytoplasm/cell). Evanalyzer measures intensities on raw channels; corrected means were approximated as mean(ch1) x mean(ch3).
About the example data
In this human cytoplasm-nucleus translocation assay, previously calculated illumination correction functions are applied to two channels, the protein content is measured in the nucleus and the cytoplasm, and the ratio is a measure of translocation.
About these images:
More about these images can be found at the BBBC (BBBC014). One 4-channel TIFF per well: channel 0 = DNA (Channel 2-*), channel 1 = protein (Channel 1-*), channels 2 and 3 = the inverse of CellProfiler’s precomputed illumination functions VitraChannel2ILLUM.npy and VitraChannel1ILLUM.npy as 8-bit images (round(255/f)), since evanalyzer cannot load .npy files.
Spot count
Count the number of spots in an image channel.
More details
Count the number of spots in an image channel.
Contributed by: Melanie Schuerz (University of Salzburg)
No downloads match the selected tags.
Modules
Section titled “Modules”.evamodule modules - a set of steps bundled into one step with its own parameters, for any pipeline. Put the file into the templates folder in EVAnalyzer’s user folder (~/.local/share/evanalyzer/templates on Linux, ~/Library/Application Support/evanalyzer/templates on macOS, %APPDATA%\evanalyzer\templates on Windows) and it shows up under Modules in the command picker of every project. See Module library.
Cell area brightfield
Cell brightfield
More details
Segments the overall cell area from a brightfield image using Sobel edge detection, blurring and an automatic threshold. The class the objects get is the module’s “Cell area” parameter.
Contributed by: Melanie Schuerz (University of Salzburg)
Cellpose-SAM
Cell segmentation with Cellpose-SAM, set up to match the Cellpose Python package
More details
Segments cells with a Cellpose-SAM model (select the exported .pt file in the module’s “Model” parameter; the model must be exported with —channels 3, see docs/convert_cellpose.py). The steps reproduce what the Cellpose Python package does: the image is stretched so the 1st percentile becomes 0 and the 99th becomes 1 (Enhance Contrast, saturated pixels 2 % = 1 % per side; computed per analysis tile, Cellpose uses the whole image), the gray image goes into all three model channels, objects are built from the flows like Cellpose (seed-based masks), holes inside each object are filled (Fill Object Holes), and objects whose shape does not match the predicted flows are removed (flow threshold 0.4). For RGB images, add a Color Filter step first to get a gray image. The Cellpose web demo additionally shrinks images to 1000 px (set ‘max resize’ to 1000) and uses 250 iterations. The last step removes very large objects (Cellpose removes objects above 40 % of the image) - set “Max area” for your images. The class the objects get is the module’s “Cells” parameter.
Contributed by: Joachim Danmayr (University of Salzburg)
EV detection
EV detection
More details
Detects extracellular vesicles (EVs) in a fluorescence channel using rolling ball background subtraction, blurring and Li thresholding. The class the objects get is the module’s “EVs” parameter.
Contributed by: Melanie Schuerz (University of Salzburg)
EV in cell detection
EV in cell detection
More details
Detects extracellular vesicles inside cells using median background subtraction, blurring, Li thresholding and watershed splitting. The class the objects get is the module’s “EVs in cells” parameter.
Contributed by: Melanie Schuerz (University of Salzburg)
Nuclei detection
Nuclei detection
More details
Detects cell nuclei using rolling ball background subtraction, blurring, Triangle thresholding and watershed splitting. The class the objects get is the module’s “Nuclei” parameter.
Contributed by: Melanie Schuerz (University of Salzburg)
Yeast cells brightfield
Yeast cells brightfield
More details
Segments yeast cells from a brightfield image using contrast enhancement, Gaussian blur, Canny edge detection, manual thresholding and a Voronoi tessellation around detected cell centers. The class the cells get is the module’s “Yeast cells” parameter. The Hough-transform cell-center detection of the original pipeline is approximated by Connected Components on the thresholded edges; the detected objects (the component’s own class “Cell centers”) are used as Voronoi centers.
Contributed by: Melanie Schuerz (University of Salzburg)
No downloads match the selected tags.
AI Models
Section titled “AI Models”Pre-trained models for the AI Segmentation and AI Object Classifier commands.
Cellpose SAM v2
AI Cellpose Segmentation
More details
Cellpose-SAM is a biological segmentation model that integrates the pretrained transformer architecture of Meta’s Segment Anything Model (SAM) with the Cellpose framework to accurately predict vector flow fields for dense cellular structures. By combining these methods, it achieves “superhuman generalization,” outperforming the average accuracy of human annotators and reaching near-optimal cell masking performance. [1]
[1]: Pachitariu, M., Rariden, M., & Stringer, C. (2025). Cellpose-SAM: superhuman generalization for cellular segmentation. bioRxiv. doi.org
Unzip the download and point Model path at the cpsam_v2.pt file inside when configuring the AI Cellpose Segmentation command.