TMARKER assists in
cell nuclei counting and
staining estimation of pathological immunohistochemical TMA images and non-TMA images. Two main scenarios are adressed with TMARKER:
- You want to know how many cell nuclei on the image are positive and negative for a certain protein. Therefore, you already did an immunohistochemical staining experiment. TMARKER provides reproducable, stable and accurate cell counting and staining estimation assistance with color deconvolution.
- For staining estimation, you only want to consider one type of cells in the image (e.g. only relevant cancer cells). TMARKER provides modern machine learning algorithms that detect these relevant cells in the image and perform staining estimation only on these cells. Also this procedure is reproducable, stable and transferable.
The software is freely available and user friendly. It is written in Java v1.7 and can be started without installation from the web. To start it, click the orange "Launch" button on top. If you don't see this button, Java might be disabled or not installed.
A
demonstration data set can be found here. It is used in the upcoming tutorial.
A new release of TMARKER is coming soon (planned Feb 2015). It will support plugins for TMA analysis.
One plugin will support computational CISH analysis on TMA images.
Please find these short tutorials for the use of TMARKER. They use the demonstration dataset.
Cell counting with color deconvolution without cell nucleus classification.
Cell nucleus classification and staining estimation (coming soon...).
Please cite TMARKER as you use it for your work:
Peter J. Schüffler, Thomas J. Fuchs, Cheng Soon Ong, Peter J. Wild, Niels J. Rupp, Joachim M. Buhmann
TMARKER: A free software toolkit for histopathological cell counting and staining estimation.
J Pathol Inform 2013, 4:2, doi 10.4103/2153-3539.109804
Go to article
If you have any questions, comments, suggestions or criticism, feel free to
contact us at peter.schueffler@inf.ethz.ch
TMARKER is open source software, made available under a BSD license. The Java source code can be downloaded here.
TMARKER comes "as is", with no warranty with respect to its fitness for purpose,
its operational state, character, quality, or freedom from defects, or the
non-infringement of rights of third parties.
Trouble Shooting
I have some technical issues in downloading the software using Win 8.
Yes, TMARKER runs also under Win 8.x (OS independent). TMARKER now runs with Java 1.7 (the most recent version),
for Java 1.7 offers smart technical details for imaging saving computational resources.
On the other hand, Java 1.7 comes with more restricted security issues.
I had to re-allow my Java system to open TMARKER over the web. In the Java settings on the system,
on the Security tab, I set the level to middle (the lowermost position).
Further, also in the security tab, I added two pages to the list of excluded web-pages
(excluded from the security borders):
http://www.comp-path.inf.ethz.ch/ and http://people.inf.ethz.ch/peschuef/. Then it worked again.
Installation
What is the difference between TMARKER Java Web Start and local installation?
There is no functional difference between TMARKER Java Web Start and local installation. The same program is just started in two different ways.
With the Java Web Start, you don't need any installation procedure, which should be most comfortable for almost all users.
If your machine doesn't support Java Web Start, or the program shows some unexpected behaviour, you can install TMARKER
on your machine. Further, the TMARKER Web Start doesn't include the demonstration dataset, which is needed if you want
to follow the tutorials or to try TMARKER out. You can also use the Web Start and download the
demonstration data separately.
How do I install TMARKER?
Please look at the top of this page and choose the installer for your platform. Currently, Windows, Linux and Mac are supported for local installation.
The Windows installer for MethMarker consists of a native .exe
file. The platform-independent MacOS / Linux / Unix installer
is a .jar
file that can be launched with the following command line statement "java -jar install_TMARKERUnix.jar
",
after the ZIP archive has been unpacked.
What are the requirements for the use of TMARKER?
- Platform-independent due to Java technology,
- Java 1.7 or higher to start the program,
- 1024 Mb RAM or more,
- if you use TMARKER for cell nuclei counting, you need tissue images in standard image formats (jpg, tiff, png, ...),
- to use TMARKER's online tools (all optional - e.g. search for updates), you need an unrestricted internet connection.
How do I run TMARKER?
The TMARKER installer will generate shortcuts (startmenu, desktop icon), which can be clicked to start the program
Alternatively, you can start TMARKER after installation with the command java -jar -Xmx1024m TMARKER.jar -d 0
in the program folder.
-d
stands for debugging level.
How do I uninstall TMARKER?
TMARKER comes with an uninstaller. You should have access to it via the shortcut "uninstall TMARKER". Alternatively, you can
directly call the uninstaller located in the program folder of TMARKER. Go to the subfolder "uninstall" and enter java -jar uninstaller.jar
.
Usage of TMARKER
What is TMARKER and when should I use it?
You can use TMARKER when you want to do one of these tasks:
- Count cell nuclei on TMA images or whole tissue slide images. These images are typically immunohistochemically stained.
- Estimate the staining percentage of IHC stained (nucleus staining) TMA images or whole tissue slide images. This task is fulfilled by
classifying the images into background and nuclei followed by classifying the nuclei into malignant and benign/other.
Therefore, you should be prepared to label the images with nuclei examples, to train the classification algorithms.
Alternatively, you can load an already trained classifier to TMARKER and use it for the staining estimation on new images.
- Perform a survival analysis on a patient cohort with TMA images. The survival information (time, event, independent covariates)
should be stored in a .csv file. See the survival tutorial for more information.
General
Can TMARKER be used offline?
Yes. All important functionalities are implemented in TMARKER. For special tasks, you can optionally use online services
(such as "search for updates") for which an unrestricted internet connection is required.
Does TMARKER support membrane staining estimation or cytoplasm staining estimation?
No, TMARKER is developed for nuclear staining estimation.
Does TMARKER support nuclear staining intensity estimation (rather than percentage estimation)?
TMARKER was originally developed for automated cell counting, cell classification and percentage staining estimation of stained malignant nuclei.
Though it is not supported, yet, we work on the extension of automated staining intensity estimation.
Does TMARKER support different types of cancer?
TMARKER is tested and developed on rencal clear cell carcinoma TMA images, prostate cancer images and mamma carcinoma images. However, every type of
cancer is supported in TMARKER, as long as similar tissue images can be produced.
Can a staining estimation classifier used on any type of cancer?
Since different type of cancers will have different tissue and image morphology, a classifier should only be trained and used on one type
of cancer. Cross-Cancer validation has not been done, yet. However, you can use a trained classifier on a new image cohort of the same
type of cancer and same staining antibody as used during training.
How do I load images in TMARKER?
Either by one of the following ways:
- Drag and drop one or more images into the big area in TMARKER.
- Click "File" -> "Open..." and select the images in the Open dialog. Make sure to have selected the right file extensions
in the Open dialog ("jpg", "png", ...). Otherwise, the image files will be hidden in the dialog.
- Drag and drop or open a .xml file which was saved in TMARKER in a previous session. XML files store the location of the images.
If the images cannot be found in the old location, you are asked to select a new location.
- Drag and drop or oben a .csv file which includes the image file names. The .csv file must be semicolon separated. One row
corresponds to one sample (image). Each column represents a certain characteristic of the sample (e.g. sample number, age of the corresponding
patient, sex of the corresponding patient, scores or other qualities of the corresponding patient). One column must contain the full filename
of the corresponding image. You will be asked to select this column. Further, the .csv should use headers in the first row.
Which image file formats are supported by TMARKER, and which should I use?
TMARKER supports all standard image formats such as JPG, JPEG, PNG, BMP.
To achieve good processing results, the images should not contain too many compression artefacts. Good JPG compression, PNG or BMP are therefore well suitable.
Which image magnification should be used for the images?
TMARKER is fairly tested on 40x TMA images. However, every magnification factor is supported. For different image magnifications,
different nuclei radii might be necessary, which can be set in the program. If you plan to apply trained classifiers on different
image sets (e.g. for automated staining estimation), the image sets should have same magnification.
Do the images must have white background? What is background correction?
The images should be scanned as neutral as possible, of course. A slightly greyish background does normally not hinder a good classfication result.
However, if you use different images with different background brightnesses, you can run TMARKER's automated background correction on selected or all
images. TMARKER will then try to find an area on the image with a greyish background and recalculate the image intensity profile such that the background
appears white. This "normalizes" the images to the same background level, such that color and intensity information is similar across the images.
What is the "output" of TMARKER? What files can I save?
TMARKER provides several possibilities to save your work:
- Save the Images as XML: This is the best way to save all annotations and parameters together with the image paths and filenames. The images itselves are not saved, but only the filenames, such that in the next session, the images should also be available on your harddisk.
- Save results as PDF: This is the best way to save the found results as PDF as a small report. This report comprises the images, annotations and found nuclei with the staining estimation.
- Save classifiers as .TMA: This is the possibility to save the nucleus detector and classifier in one TMA file (a new binary file format which captures java objects). These classifiers can be loaded by other TMARKER instances.
Staining Estimation
How does TMARKER's staining estimation work?
In principle, it works like this:
- You load images into TMARKER that you want to estimate.
- You start the "Semisupervised Labeling" process (hidden in the "Superpixels..." dialog). TMARKER will then run some image preprocessings
on the images (like superpixel creation, feature extraction, ...).
- You provide the first labels to TMARKER (e.g. label maligne cells nuclei, benign nuclei and background).
- During labeling, TMARKER suggests other nuclei that has been classified on the images. Also, TMARKER updates the staining estimation
for the images, under the condition that the suuggested nuclei are right.
- You continuously supervise the classifier's guess by correcting or adding additional labels, until you are satisfied with the result.
TMARKER will steadily update the classification guesses and staining estimation for the images.
- After you have finished the labeling, you can save the classifier and use it for further staining estimations on different images.
Image Labeling
Which types of annotations can I provide?
Two types of annotations are supported in TMARKER:
- Single nucleus annotations: You can label on the images the single cell nuclei by click on the images. In the "Nucleus Annotation", you can select the type of nucleus that you add to image by mouseclick: cancerous nucleus (red), other/benign nucleus (green) or background (white). The cancerous and benign nuclei are further separated in staining intensity 0-3, for differently intensively stained nuclei, if applicable. These annotations will be used for the supervised learning steps in the staining estimation process.
- Whole Image annotations: In the "Whole Image Annotation", you can grade the whole image, e.g. for the overall percentage score, the overall intensity score, a comment or a different annotation. This will be stored as property of the image. Such annotations can be used for comparison reasons (e.g. "man vs machine").
Both annotations will be saved, if you store the (selected) annotated images as .XML.
Cell Counting with color deconvolution
How do I find good parameters for the color deconvolution?
This is the tricky part for the method. In principle, the parameters depend on the images itself (staining intensity, tissue morphology, ...).
When you have selected the right staining type (e.g. "H DAB"), the parameter "t_dab" and "t_hema" should be found very quickly, by try and error.
You typically would process the color deconvolution on one image and find the best parameters which find the right nuclei on this image. Then, you select all images and run the color deconvolution with these found parameters. Color deconvolution is always processed only on seleted images.
Survival Analysis