Bioimage Analysis

Pre-processing:

1. • Crop: Used to reduce background/no-information areas. Reduces the size of an image.
   • Invert: Reverses the intensity scale of an image so that black is white and vice versa. Algorithms may need white objects on a black background instead of vice versa.
   • Filtering: Used to reduce noise or smooth edges of objects.  It can improve contrast to differentiate objects from non-objects/background.
   • Color Threshold: Used to extract or separate features based on color (for example when analyzing histology images).
   • Illumination Correction: Can be used to correct for irregular/uneven illumination due to optical instrumentation, etc.
   • Registration: Used to align images, that is, transform different sets of image data into one coordinate system.
   • Stitching: Used to reconstruct large images/stacks from an arbitrary number of individual images/stacks.
   • Deconvolution: Used to correct the systematic error – or ‘blur’ – within an image introduced by optical systems.
2. Often this is a result of uneven illumination due to the imaging system used.  This needs to be corrected using illumination correction techniques to improve the overall signal in the stitched image, especially for subsequent analysis steps, including segmentation, etc.
3. Deconvolution is the process of removing systematic error – or ‘blur’ – introduced in acquired images by the optical system.  The Point Spread Function describes the response of the imaging system to a point-like object; it is a measure of the system’s convolution.

Segmentation:

1. Segmentation is the process of identifying individual objects within an image.  It can be used to measure individual biological entities such as individual synapses, nuclei, cells, vesicles, regions of an entire organism or tissue, or cells growing in culture.
2. 1. Expand (dilate)
   2. Shrink (erode)
   3. Skeletonize (convert to a single pixel width)
   4. Subtract (mask one area with another)
   5. Relate (associate objects with each other)
   6. Merge (connect objects together)
3. 1. Brightfield/DIC/phase contrast
   2. Tissue images, including histopathology
4. 1. Pros: powerful, doesn’t need advanced image processing knowledge to use them
   2. Cons: requires labeling/annotating a lot of examples, may need to re-train on each experimental batch

Tracking:

1. 1. Segmentation: identifying objects of interest
   2. Linking: connecting object-to-object from frame-to-frame
   3. Measure: object & track statistics, cell lineage, etc.
2. 1. Following cells as they move over time to better understand their speed/homing.
   2. Tracking vesicles moving within a cell to assess their directionality.
   3. Tracking dividing cells to follow cellular lineages.

Measurement:

1. 1. Spatial Relationships
   2. Texture (smoothness)
   3. Intensity
   4. Counts
   5. Size
   6. Shape
2. False. While corrections can indeed be performed, it’s best to improve acquisition as much as possible.
3. A classifier is a set of rules that a computer uses to predict the correct classification (of an individual pixel or of an entire biological entity such as a cell). It requires the biologist pointing out (annotating) many example images of each phenotype so that it can learn to distinguish phenotypes from each other.

Best Practices:

1. 1. Keep raw data
   2. Document your workflows
   3. Safely store your data