Module: MeasureImageQuality

This option lets you choose which images will have quality metrics calculated.

• All loaded images: Use all images loaded with the Input modules. The quality metrics selected below will be applied to all loaded images.
• Select...: Select the desired images from a list. The quality metric settings selected will be applied to all these images. Additional lists can be added with separate settings.

(Used only if "Select..." is chosen for selecting images)
Choose one or more images from this list. You can select multiple images by clicking using the shift or command keys. In addition to loaded images, the list includes the images that were created by prior modules.

Select Yes to add the image's rescaling value as a quality control metric. This value is set only for images that loaded using the Input modules. This is useful in confirming that all images are rescaled by the same value, since some acquisition device vendors may output this value differently. See NamesAndTypes for more information.

Select Yes to compute a series of blur metrics. The blur metrics are the following, along with recomendations on their use:

• PowerLogLogSlope: The power spectrum contains the frequency information of the image, and the slope gives a measure of image blur. A higher slope indicates more lower frequency components, and hence more blur (Field, 1997). This metric is recommended for blur detection in most cases.
• Correlation: This is a measure of the image spatial intensity distribution computed across sub-regions of an image for a given spatial scale (Haralick, 1973). If an image is blurred, the correlation between neighboring pixels becomes high, producing a high correlation value. A similar approach was found to give optimal performance for fluorescence microscopy applications (Vollath, 1987).
  Some care is required in selecting an appropriate spatial scale because differences in the spatial scale capture various features: moderate scales capture the blurring of intracellular features better than small scales and larger scales are more likely to reflect intercellular confluence than focal blur. A spatial scale no bigger than the feature of interest is recommended, although you can select as many scales as desired.
• FocusScore: This score is calculated using a normalized variance, which was the best-ranking algorithm for brightfield, phase contrast, and DIC images (Sun, 2004). Higher focus scores correspond to lower bluriness.
  More specifically, the focus score computes the intensity variance of the entire image divided by mean image intensity. Since it is tailored for autofocusing applications (difference focus for the same field of view), it assumes that the overall intensity and the number of objects in the image is constant, making it less useful for comparision images of different fields of view. For distinguishing extremely blurry images, however, it performs well.
• LocalFocusScore: A local version of the Focus Score, it subdivides the image into non-overlapping tiles, computes the normalized variance for each, and takes the mean of these values as the final metric. It is potentially more useful for comparing focus between images of different fields of view, but is subject to the same caveats as the Focus Score. It can be useful in differentiating good versus badly segmented images in the cases when badly segmented images usually contain no cell objects with high background noise.

References

• Field DJ (1997) "Relations between the statistics of natural images and the response properties of cortical cells" Journal of the Optical Society of America. A, Optics, image science, and vision, 4(12):2379-94. <(pdf)
• Haralick RM (1979) "Statistical and structural approaches to texture" Proc. IEEE, 67(5):786-804. (link)
• Vollath D (1987) "Automatic focusing by correlative methods" Journal of Microscopy 147(3):279-288. (link)
• Sun Y, Duthaler S, Nelson B (2004) "Autofocusing in computer microscopy: Selecting the optimal focus algorithm" Microscopy Research and Technique, 65:139-149 (link)

(Used only if blur measurements are to be calculated)
The LocalFocusScore is measured within an N × N pixel window applied to the image, whereas the Correlation of a pixel is measured with repsect to its neighbors N pixels away.

A higher number for the window size measures larger patterns of image blur whereas smaller numbers measure more localized patterns of blur. We suggest selecting a window size that is on the order of the feature of interest (e.g., the object diameter). You can measure these metrics for multiple window sizes by selecting additional scales for each image.

Select Yes to calculate the saturation metrics PercentMaximal and PercentMinimal, i.e., the percentage of pixels at the upper or lower limit of each individual image.

For this calculation, the hard limits of 0 and 1 are not used because images often have undergone some kind of transformation such that no pixels ever reach the absolute maximum or minimum of the image format. Given the noise typical in images, both these measures should be a low percentage but if the images were saturated during imaging, a higher than usual PercentMaximal will be observed, and if there are no objects, the PercentMinimal value will increase.

Select Yes to calculate image-based intensity measures, namely the mean, maximum, minimum, standard deviation and median absolute deviation of pixel intensities. These measures are identical to those calculated by MeasureImageIntensity.

Automatically calculate a suggested threshold for each image. One indicator of image quality is that these threshold values lie within a typical range. Outlier images with high or low thresholds often contain artifacts.

(Used only if image thresholds are calculcated)
Select Yes to calculate thresholds using all the available methods. Only the global methods are used.
While most methods are straightfoward, some methods have additional parameters that require special handling:

• Otsu: Thresholds for all combinations of class number, minimzation parameter and middle class assignment are computed.
• Mixture of Gaussians (MoG): Thresholds for image coverage fractions of 0.05, 0.25, 0.75 and 0.95 are computed.

See the IdentifyPrimaryObjects module for more information on thresholding methods.

(Used only if particular thresholds are to be calculated)
This setting allows you to apply automatic thresholding methods used in the Identify modules. Only the global methods are applied. For more help on thresholding, see the Identify modules.

(Used only if threshold are calculated and MoG thresholding is chosen)
Enter the approximate fraction of the typical image in the set that is covered by objects.

(Used only if thresholds are calculcated and the Otsu thresholding method is used)
Select Two classes if the grayscale levels are readily distinguishable into foregound (i.e., objects) and background. Select Three classes if there is a middle set of grayscale levels that belongs to neither the foreground nor background.

For example, three-class thresholding may be useful for images in which you have nuclear staining along with a low-intensity non-specific cell staining. Where two-class thresholding might incorrectly assign this intemediate staining to the nuclei objects, three-class thresholding allows you to assign it to the foreground or background as desired. However, in extreme cases where either there are almost no objects or the entire field of view is covered with objects, three-class thresholding may perform worse than two-class.

(Used only if thresholds are calculcated and the Otsu thresholding method with Three classes is used)
Choose whether you want the middle grayscale intensities to be assigned to the foreground pixels or the background pixels.