main:imageselectionmodel
Differences
This shows you the differences between two versions of the page.
| Both sides previous revision Previous revision Next revision | Previous revision | ||
|
main:imageselectionmodel [2017/08/03 20:32] bshirley |
main:imageselectionmodel [2017/11/09 18:44] (current) yli [Contents of an image selection model] |
||
|---|---|---|---|
| Line 1: | Line 1: | ||
| ====== Image selection model ====== | ====== Image selection model ====== | ||
| - | An image selection model is a series of filters which exclude undesirable images within a sample from use in [[main: | + | An image selection model is a series of filters which exclude undesirable images within a sample from use in [[main: |
| ===== Apply an image selection model ===== | ===== Apply an image selection model ===== | ||
| Line 21: | Line 21: | ||
| * **Obj - Exclude if object count is < [value] or > [value]**\\ The number of total objects found in an image, excluding noises and nuclei. Ideally, it would be 46. This filter excludes images with too few or too many chromosomes. | * **Obj - Exclude if object count is < [value] or > [value]**\\ The number of total objects found in an image, excluding noises and nuclei. Ideally, it would be 46. This filter excludes images with too few or too many chromosomes. | ||
| * **Seg - Exclude if segmented count < [value] or > [value]**\\ The number of objects processed by GVF algorithm in an image. Segmented objects (processed by GVF) are subset of total objects. Ideally, it would also be 46. This filter is similar to object count filter, but more stringent. | * **Seg - Exclude if segmented count < [value] or > [value]**\\ The number of objects processed by GVF algorithm in an image. Segmented objects (processed by GVF) are subset of total objects. Ideally, it would also be 46. This filter is similar to object count filter, but more stringent. | ||
| - | * **Ratio - Exclude if classified object ratio > [value]**\\ The ratio of objects recognized as chromosomes and segmented objects. Classified objects (recognized as chromosomes) are subset of segmented objects. This filter excludes images in which chromosomes are not recognized effectively. | + | * **Ratio - Exclude if classified object ratio < [value]**\\ The ratio of objects recognized as chromosomes and segmented objects. Classified objects (recognized as chromosomes) are subset of segmented objects. This filter excludes images in which chromosomes are not recognized effectively. |
| * **Score - Image ranking and inclusion method**\\ We recommend the minimum number of images remaining after image selection to be greater than 200-250 images.((Liu J, Li Y, Wilkins R, Flegal F, Knoll JHM, Rogan PK. Accurate Cytogenetic Biodosimetry Through Automation Of Dicentric Chromosome Curation And Metaphase Cell Selection. bioRxiv 120410; doi: 10.1101/ | * **Score - Image ranking and inclusion method**\\ We recommend the minimum number of images remaining after image selection to be greater than 200-250 images.((Liu J, Li Y, Wilkins R, Flegal F, Knoll JHM, Rogan PK. Accurate Cytogenetic Biodosimetry Through Automation Of Dicentric Chromosome Curation And Metaphase Cell Selection. bioRxiv 120410; doi: 10.1101/ | ||
| * **None**: ADCI uses no image ranking and inclusion. | * **None**: ADCI uses no image ranking and inclusion. | ||
| Line 40: | Line 40: | ||
| 3. To create new Image Exclusion Filter thresholds: Examine distributions to find thresholds (as +S.D. above or below the mean value for the test) that can be applied to the majority of samples that remove a subset of images. This should be based on outlier values that distinguish the majority of images for a test from rarer images. For example, we expect 46 chromosomes in a metaphase cell. Most images will contain <60 objects (chromosomes + nuclei). Images with >60 objects are infrequent and are more likely to indicate excessive sister chromatid separation or multiple metaphase cells, or a lot of debris may be present. | 3. To create new Image Exclusion Filter thresholds: Examine distributions to find thresholds (as +S.D. above or below the mean value for the test) that can be applied to the majority of samples that remove a subset of images. This should be based on outlier values that distinguish the majority of images for a test from rarer images. For example, we expect 46 chromosomes in a metaphase cell. Most images will contain <60 objects (chromosomes + nuclei). Images with >60 objects are infrequent and are more likely to indicate excessive sister chromatid separation or multiple metaphase cells, or a lot of debris may be present. | ||
| - | 4. To test a threshold for a test, open the metaphase viewer for each sample, click on “apply image filters”, and check the box corresponding to the threshold you wish to apply, then change the default threshold value(s) to those you determined using the Sample Report. The number of images in the subset will usually be reduced from the complete set of images. Select view excluded images and review these. Verify that you want to remove these images by sorting the list by, for example, object count (top right selection), and carefully reviewing images close to the threshold boundary that you have chosen. You can scroll through the list using the green arrows (bottom left), or jump to particular images (threshold is shown in brackets) using the file name selector. If necessary, adjust the threshold and repeat the filtering and review step until you have selected a threshold. Record this threshold and the associated test. | + | 4. To test a threshold for a test, open the [[main: |
| 5. Repeat step 4 for the other tests that you wish to apply to the data. The filtering is cumulative: images that are removed by one test, if selected for removal by another, will not be repeatedly eliminated. | 5. Repeat step 4 for the other tests that you wish to apply to the data. The filtering is cumulative: images that are removed by one test, if selected for removal by another, will not be repeatedly eliminated. | ||
| - | 6. After all of the image selection filters have been determined, then decide whether additional thresholds are required based on sorting. The combined Z score or Group Bin methods can sort images from highest to lowest quality, then the top X images can be selected for dose estimation or calibration curve generation. X should be at least 250, however if a larger number of high quality images pass the preceding tests (steps 4 and 5), then the top scoring image count can be higher (≥ 500). | + | 6. After all of the image selection filters have been determined, then decide whether additional thresholds are required based on sorting. The combined Z score or [[main: |
| 7. Save the combined image selection model on the “apply image selection model” window by selecting “Save Current User Customized Model” and assigning a name to the model. Hint: assign a model name that describes the filters used in the model for future reference. The model name will appear in you list of models that can be applied to future samples. | 7. Save the combined image selection model on the “apply image selection model” window by selecting “Save Current User Customized Model” and assigning a name to the model. Hint: assign a model name that describes the filters used in the model for future reference. The model name will appear in you list of models that can be applied to future samples. | ||
| Line 50: | Line 50: | ||
| 8. Note: After Z score based thresholding, | 8. Note: After Z score based thresholding, | ||
| + | 9. [[main: | ||
main/imageselectionmodel.1501792375.txt.gz · Last modified: 2017/08/03 20:32 by bshirley
Page Tools
