«IHC Membrane Image Analysis User’s Guide ©Copyright 2007 Aperio Technologies, Inc. Part Number/Revision: MAN-0026, Revision B Date: January 2, ...»
There are different ways to access the digital slides over a file system (local or mapped drive) for the parameter tuning. In the case that the images folder where the digital slides are stored on the ImageServer is accessible over the network, those digital slides can be accessed directly by ImageScope over the file system. If this is not possible, then you need to place a copy of the digital slides somewhere where you can access them over the file system. Another possibility to create a data set of digital slides for algorithm tuning is to extract the significant regions from the digital slides (use the annotations from the pathologists to identify those regions and their significance) and store them locally.
To extract a region in ImageScope:
1. Click on the Extract Region icon.
2. Outline the region to extract.
3. Provide the output file location and filename in the Extract Region Image Region window and click Extract.
To open a digital slide over the file system with ImageScope, either: Double-click on the file and it will open with ImageScope or in ImageScope select Open Image… in the File drop-down menu, browse to the digital slide and click Open.
The annotations from the pathologists provide guidance on which regions should be analyzed and what their significance is. Because the annotations are stored in Spectrum Information Manager’s database, we need to export those annotations into the same local folder where the digital slides svs files are stored.
The annotation xml files should be saved with the same filename as their corresponding svs files but with an xml file extension. When a digital slide svs file and annotation xml file have the same filename, ImageScope automatically loads the corresponding annotation file when opening the digital slide.
To find the corresponding svs file for a digital slide in Spectrum Information Manager, you can look up its filename originally provided by the ScanScope in ImageScope’s Image Information window opened by clicking on the information icon. Note that as the digital slides were imported from a folder, the digital slides are in inverse order in Spectrum Information Manager as the digital slides in the folder. To save the annotations to a file, bring up the Annotations window in ImageScope by selecting Annotations Window from the View drop-down menu, click on the export annotation to file icon and provide the filename and location.
When opening digital slides over the file system and running algorithms, the location where the markup images are stored is determined by ImageScope. If multiple users (multiple ImageScopes) want to see the markup images, then we need to create a common markup image folder accessible by everybody and specify this folder in all their ImageScopes. To specify the markup image folder in ImageScope, select Options… in the Tool drop-down menu, select the General tab and enter the Markup Image Folder under Analysis.
Also, make sure that the person responsible for the algorithm parameter tuning has the same version of the algorithm installed on his/her computer as is installed on the ImageServer!
Running the Algorithm and Adjusting the Parameter Settings Now when the digital slides are opened over the file system in ImageScope, the algorithm parameters can be adjusted and the results using different parameter settings can be evaluated quickly for a couple of digital slides.
1. First, bring up the Algorithms window by selecting Analysis in the View drop-down menu.
2. Then select the IHC Membrane Algorithm by clicking Select Algorithm, selecting the IHC Membrane Algorithm, and clicking Select.
3. Now the selected algorithm is loaded and its parameters are shown in the Algorithms window where its settings can be adjusted. Parameters can be modified by changing their values or selecting a different choice in the drop-down menus. Note that new parameter values only take effect after the cursor has been moved off the parameter value—once you have modified a parameter value, just click on a different parameter.
4. To use the algorithm for parameter tuning, set Use Mode to 0 – Analysis/Tuning. and Markup Image Type to 0 – Tuning.
When the parameter settings are modified, ImageScope will make them the default parameter settings as long as the algorithm is loaded. If the same or another algorithm is selected, the factory default settings are applied and the current parameter settings are lost. To save a parameter setting you can export the algorithm with its parameter setting as a macro to a file by clicking Export Macro… in the Algorithms window, providing the file location and filename and clicking Save.
You may want to save a couple of good parameter settings to be able to evaluate them against each other over the entire training data set. It might be a good idea to use a filename that identifies the stain, the manufacturer and the tissue type as well as the Tuning algorithm mode
and a parameter setting identifier. Store the algorithm macros with the training data set so you know what they belong to and can find them easily.
The user has a choice between Current Screen, the Entire Image and the regions outlined in the Selected Annotation Layer for the Region of Analysis. Use caution when using Current Screen as the digital slide will be processed at the resolution defined by the zoom level in ImageScope and will only be processed at 20X if the zoom level is set to 20X (required for the IHC Membrane Algorithm). To evaluate different parameter settings, the best approach is to select a small region that exhibits certain characteristics to which the algorithm should be tuned and select Selected Annotation Layer and then study the impact of changing the parameter settings. Note that the processing of the Entire Image takes a long time and does not necessarily provide more valuable data for the algorithm tuning.
5. The best way to evaluate the results of an image analysis algorithm is visually; therefore select the Generate Markup Image check box.
The markup image for tuning shows the nuclei in green with a black outline, the cells outlines or membranes as black outlines and the membrane staining in red. Note that the scoring scheme parameters that determine the classification of the cells have not been determined yet therefore showing the color-coded classification of the cells would not be meaningful.
6. To run the algorithm, click Run. The algorithm will return its results in a new layer in the Annotations window. The progress and completion of the analysis is shown at the bottom of the Algorithms window.
The Annotations window provides a powerful tool for managing the different results provided by the algorithm for different parameter settings.
For example, we used Layer 1 to outline the regions of analysis and ran the algorithm first with the default settings and then four times with different values (0, 1, 2 and 5) for the Averaging Radius (um) parameter creating thereby four child layers with different algorithm results. The algorithm parameters used to run the algorithm are always stored with the algorithm results under Layer Attributes, making it
32 IHC Membrane Image Analysis User’s GuideChapter 6 – Algorithm Setup
easy to remember the exact parameter settings. Comparing the markup images on the same regions and outputs from different parameter settings is as easy as clicking on the different layers. To review the original image, click on the show/hide layer icon in the Layers section of the Annotations window and toggle between the markup image and the original image.
Test Parameter Settings To test one or more parameter settings on the entire training data set, import the algorithm macros into Spectrum Information Manager and run them in batch mode on all digital slides in the training data set using the annotations provided by the pathologists to define the regions of analysis.
1. Import an algorithm macro into Spectrum Information Manager by first opening an Internet browser and logging into Spectrum Information Manager using the same workstation used to save the tuning parameters and going to the Analysis Macros view by clicking Macros under the Analysis drop-down menu.
2. Then in the Analysis Macros view, click Add Macro, provide the tuned parameter set location and filename and a Macro Name to identify this macro by in Spectrum Information Manager and click Save. It might be a good idea to use a Macro Name that identifies the stain, the manufacturer and the tissue type as well as the Tuning algorithm mode and a parameter setting identifier.
3. Navigate in Spectrum Information Manager to the training data sets by clicking All Cases under the Cases drop-down menu and then opening the case (project) for the specific application by clicking on the corresponding open data icon. Go to the different training data sets by opening the corresponding specimen (data set) by clicking on the open data icon.
4. To run an algorithm macro in batch mode on all digital slides grouped under a specimen (data set) first select all digital slides under Specimen Digital Slides by selecting the check box in the title bar, and then click Analyze.
5. In the Analyze Images view, select the algorithm macro in the Select Analysis Macro drop-down menu, select the annotation layer that contains the annotations from the pathologists with the tumor-cell only region outlines to be processed in the Select Input Annotation Layer drop-down menu, and click Analyze.
The Pending Analysis Jobs window then shows the progress on the batch processing. To update the view, click on the Internet Explorer refresh icon.
6. To review the results from the different parameter settings, go back to the training data sets, open all digital slides with ImageScope and compare the markup images of all the regions and algorithm outputs from the different parameter settings. Determine the best parameter setting.
7. Delete the parameter tuning algorithm macros in Spectrum Information Manager if they are no longer required. To delete an algorithm macro, go to the Analysis Macros view by clicking Macros on the Analysis menu. Then delete a specific macro by clicking Delete after the Macro Name.
Scoring Scheme Parameters The scoring scheme parameters specify the staining intensity and membrane completeness thresholds that determine the individual tumor cell classification. The scoring scheme parameters can be adjusted
either manually like the cell feature detection parameters or trained automatically by presenting the algorithm with a set of digital slides that have been scored according to a scoring standard. The automatic training evaluates all possible threshold combinations to determine the thresholds that provide scores that best match the established standard scores.
Parameter List Cell classification thresholds Weak(1+) Threshold: Membrane intensity threshold for weak (1+) [0 Value ≤ 255] (Default: 200).
Moderate(2+) Threshold: Membrane intensity threshold for moderate (2+) [0 Value ≤ 255] (Default: 175).
Strong(3+) Threshold: Membrane intensity threshold for strong (3+) [0 Value ≤ 255] (Default:
Completeness Threshold: Threshold for classifying membrane as completely stained [0 Value 100] (Default: 80).
Algorithm Modes The IHC Membrane Image Analysis algorithm has three different modes of operation: (1) Tuning, (2) Training and (3) Analysis with their respective markup images to choose from. Training the scoring scheme parameters is done in training mode.
Use Mode: Select Training or Analysis/Tuning Mode [0 – Analysis/Tuning, 1- Training] (Default:
0 – Analysis/Tuning).
Mark-up Image Type: Choose type of markup image. [0 – Tuning, 1 – Analysis, 2- Analysis (With Cell Edges)] (Default: 1 – Analysis).
Automatic Parameter Training Classifier Type: Select type of classifier [0 – IHCMembrane, 1- IHCNuclear] (Default: 0 – IHCMembrane).
Classifier Definition File: Name of classifier definition file [folder name]. (Default:
Automatic Parameter Training Create an algorithm macro for the automatic parameter training.
Start with the cell feature detection parameters that yield the best results on the training data sets.
1. Open ImageScope (just the program) and import the algorithm macro with the best cell feature detection parameter settings.
2. To import an algorithm macro, first click Analysis on the View menu to show the Analysis window.
3. In the Analysis window, click Import Macro… and provide the file location and filename of the algorithm macro and click Open.
4. To use the algorithm for automatic training, set the Use Mode to 1 – Training.
5. Set the Classifier Definition File parameter to the name of the folder to receive the training results. The algorithm will create a folder with this name in the ScanScope program folder under AAF and Algorithms (\Program Files\ScanScope\AAF\Algorithms). As we will run the algorithm macro on the server, the folder will be created on the ImageServer. Each time you want to automatically train the scoring scheme parameters, you need to provide a different Classifier Definition File folder name. You may want to choose a name that identifies the stain, the manufacturer and tissue type as you need to train the scoring scheme parameters for each specific application. Note that each time you run the algorithm on a digital slide in training mode and use an existing Classifier Definition File folder, it will add to the training and update the trained scoring scheme parameters.
a) To re-start/re-initialize the training, delete the Classifier Definition File folder.
b) To finish the training, change the User Mode to 0 – Analysis/Tuning and the training results will no longer be updated.