Working with Optical Flow Images

Outputs of the Optical Flow algorithm

• Perform Optical Flow acquisition and loading as detailed here.
• Output as Absolute value of vectors: pixels indicate the magnitude of velocity, but not the direction.
• Output as X and Y components of vectors: results a pair of Image Windows containing the x and y components of the velocity vectors (because pixel values can be only scalar) 
• Output as Absolute value of Projected Vectors: Positive velocities indicate movement away from the center point ROI, negative velocities indicate movement towards the ROI. To set up the center ROI when using a the Multi-Dimensional Open dialog, first load raw images by setting the Processing panel to None in the Open tab. Draw ROI on the opened image. Set the ROI No. in the Projection ROI parameter. The ROIs are automatically copied from the last open image during Optical Flow open. Perform Optical Flow Open.

Determining mean absolute velocities of mitochondria in whole cells

• The example below given on the Metamorph example data set (download from here).
• Perform Optical Flow acquisition and loading as detailed here, by setting the Output as Absolute value of vectors parameter Yes, and the other Outputs No.
• The non-fluorescent (non-edge) parts of the image are already masked by the Optical Flow algorithm. Use in the toolbar to draw a large ROI including the whole cell (dendrites of the hippocampal neuron in the example below) excluding the soma (optionally).
• Press in the toolbar to plot mean pixel values, which is absolute velocity in mm/pixel (as defined by the Pixel size parameter)
• In some cases additional masking may be required, e.g. to exclude fast moving 'jumping between frames' mitochondria that does not result an accurate Optical Flow. In this case masks are calculated from minimal intensity projections of the short time lapses. See masking below.

Neuron encircled without the soma	Mean absolute velocity diagram The y-axis is scaled in mm/sec.

Processing other channels of the Multi-DimensionalRecording

To obtain graphs of fluorescence intensities in other channels recorded in each measurement cycle preceding the short time lapse of the Optical Flow recording:

1. Load Optical Flow as above.
2. Set Processing in the Multi-Dimensional Open dialog to None.
3. Check all channels (including the (first) Optical Flow channel)
4. Press Open
5. In the main menu click (Set image windows linkage)
6. In the Set image windows linkage dialog check the name of the Optical Flow Image Window. Press OK.
7. The ROI previously drawn on the Optical Flow image now appears in the newly loaded Image Windows. To obtain fluorescence intensities:
8. First perform background subtraction.
9. Optionally, perform masking of fluorescence image channels to look only the same pixels as the Optical Flow image does (see below).
10. Plot each image window by pressing in the toolbar.

Masking of Optical Flow and other fluorescence channels

See more about masking here.

Masking using the Optical Flow image:

1. Copy (Duplicate, linked) the Optical Flow image. Work on the copy below.
2. Select the Threshold function, in the main menu/Segmentation. The thresholding is set to give 1 for any numerical value in the image. Mask values automatically result 0.
   1. Threshold value calculation method: Pixel Value
   2. Way: Above
   3. Value: If absolute velocity was calculated give -1. If negative velocities occurr, give a smaller negative number, e.g. -1000
   4. Threshold from local max/min: None (the value of Determine boundaries at does not apply for None)
   5. Process e.g. by using the context menu of the Image Window.
3. Use the Math/Image Arithmetic function to divide each background subtracted fluorescence channel by the binarized Optical Flow image. Division by zero will create the mask.
   1. Type: /
   2. Select all of the fluorescence channels as Image A, and the binarized copy as Image B. Press in the tool bar. Continue working with the results.

Optical Flow converted to mask. (Gray scale LUT was applied after the Threshold)	High pass filtered and masked TMRM channel	TMRM intensity plot

Using a mask created by locally adaptive thersholding from a fluorescence image:

1. Copy (Duplicate, linked) the non-processed fluorescence image. Work on the copy below.
2. If working with wide-field microscopy, perform high pass filtering as local background removal to amplify mitochondrial details.
   1. Select the 2D DFT Butterworth BP filter function, in the main menu/Filters. Typical parameters are (in cycles/mm; for this image resoultion have to be present in the file or set in the Setup DFT Filter):
   2. Cut On ω: 0.3 (increase this value to make a stronger background supression, or decrease if the result is too thin, noisy)
   3. Cut On order: 1.3
   4. Cut Off ω: 2000 (this is off the scale, so the filter is a high pass filter)
   5. Cut Off order: 100
   6. unit of ω: cycles/um
   7. Filter normalization: Corrected Integral
   8. Preserve edges: Yes
   9. Enlarge paper: No
   10. Enlarge to 2^: this value does not matter if above is No
   11. Leave only phase: No
   12. Absolute: No
   13. Protect MASK: No
   14. Process e.g. using the context menu of the Image Window.
3. Smooth image with Wiener filtering:
   1. Select the Wiener filter function, in the main menu/Filters.
   2. Mask width: 3
   3. Noise level: 0.0005 (increase this value to increase smoothing)
   4. Process e.g. by using the context menu of the Image Window.
4. Remove values below zero using the bottom functionality of the Threshold function:
   1. Select the Threshold function, in the main menu/Segmentation
   2. Threshold value calculation method: Pixel Value
   3. Way: Bottom
   4. Value: 0
   5. Threshold from local max/min: None (the value of Determine boundaries at does not apply for None)
   6. Process e.g.by using the context menu of the Image Window.
5. Perform locally adaptive binarization by the Threshold function:
   1. Select the Threshold function, in the main menu/Segmentation
   2. Threshold value calculation method: Otsu by series
   3. Way: Above
   4. Value: 0.5 (decrease this value to have more mitochondria or increase to have less interference by background noise)
   5. Determine boundaries at: 10 (% from the local maxium for each mitochondrion. Increase this value to get smaller mitochondria)
   6. Threshold from local max/min: Bound Maxima Locally (each object brighter than the local background by 0.5*Otsu optimal threshold will be marked down to its 10% of intensity. Note, that if using high pass filtering, 0% is the full width half max.)
   7. Process e.g. by using the context menu of the Image Window.
6. Use the Math/Image Arithmetic function to divide each background subtracted fluorescence channel by the binarized Optical Flow image. Division by zero will create the mask.
   1. Type: /
   2. Select all of the fluorescence channels and the Optical Flow image as Image A, and the binarized copy as Image B. Press in the tool bar. Continue working with the results.

Note: it is worthwhile to use the Pipeline for such complex processing.

Non-processed, projection image of the short time lapse.	High pass filtered...	Wiener filtered	Locally adaptive binarized
TMRM channel high pass filtered with similar parameters as above, but with Absolute: Yes Shown with 'Fire' LUT	Masked TMRM image, shown with Pseudocolor LUT	(double) masked Optical Flow Image. Same frame as shown in the above examples.

Determining mean radial velocities of mitochondria in whole cells

• Mean radial velocities carry the information whether anterograde and retrograde transport is in balance
• Before performing Optical Flow load the non-processed image sequence by setting the Processing in the Multi-Dimensional Open dialog to None.
• Use the point ROI tool to mark the center of the cell.
• In the Multi-Dimensional Open dialog Processing set Optical Flow. In the Optical Flow panel set the Projection ROI parameter as number of the point ROI.
• Set Output as Absolute value of Projected Vectors: Yes
• Press Open.

Non-processed image with the point ROI in the soma	Radial velocity image (note the scaling in the bottom)	Mean radial velocity diagram (press Active in the status bar of the Plot Window to show only the trace of the active ROI). The y-axis is scaled in mm/sec.

Determining velocities of individual mitochondria

Velocity measurement of individual mitochondria is based on segmentation of the same (non-processed) images that is used for the calculation of the Optical Flow. While segmentation is discussed elsewhere in detail. Here we provide an example for robust segmentation of 'mitochondrial' images. More details here later...

Using Optical Flow in Pipeline

More details here later...