Inconsistent segmentation between Cell-ID run via Rcell2 and VCell-ID

[gaspanico]

Segmentation results are not identical between Cell-ID run from Rcell2 and Cell-ID run with VCell-ID, despite choosing the same input parameters.

• Left: Segmentation with cellid run with the Rcell2 cellid wrapper with the cellID-linux compilation.
• Middle: Segmentation with VCell-ID using an "original" (unmodified) cellid compilation.
• Right: Segmentation with VCell-ID using the cellID-linux compilation.

Note how cell #304 is split from its mother cell in the left image, whereas it is connected in the two rightmost images.

A possible but not yet confirmed cause may be parameter formatting, and resulting rounding errors:

• Left: Example parameters passed to cell via the Rcell2 cellid wrapper. This is the format suggested by Andrew Gordon in the Cell-ID manual.
• Right: Parameters passed to cell via VCell-ID

Note the commas in the VCell-ID parameters instead of the standard period decimal points.

The reason I think this may be causing a formatting and/or rounding error is seen in this screen capture:

By setting the input parameters in VCell-ID as show above (max dist over waist and max split over mayor axis; tracking comparison is irrelevant here), segmentation results in the same split as that seen using the Rcell2 wrapper.

[naikymen]

Comparison bewteen VCellID, CellID-mask_mod (CLI) and CellID-mask_mod from Rcell2.

Summary: parameters seem to be passed homogenously to CellID. However using a comma as decimal separator in a parameter set to 0,8, caused to be read as 0 (zero).

VCellid Run

 max_split_over_minor 0.50
 max_dist_over_waist 8.00
 max_pixels_per_cell 1500
 min_pixels_per_cell 75
 background_reject_factor 1.00
 tracking_comparison 0.20
 align_fl_to_bf
 image_type brightfield
 bf_fl_mapping list

Using parameters:
   max_dist_over_minor_axis=5.000000e-01
   max_dist_over_waist=8.000000e+00
   back_reject=1.000000e+00
   max_pixels_per_cell=1500
   min_pixels_per_cell=75
   I_over_U_for_match (tracking parameter)=2.000000e-01

Run from CLI

Using the same parameters.txt file

../cell -b /home/nicomic/Software/cellID-linux/test_data_tmp/Position01/bf_vcellid.txt -f /home/nicomic/Software/cellID-linux/test_data_tmp/Position01/fl_vcellid.txt -o /home/nicomic/Software/cellID-linux/test_data_tmp/Position01/out -p /home/nicomic/Software/cellID-linux/test_data_tmp/Position01/parameters_vcellid_out.txt -m

Using parameters:
   max_dist_over_minor_axis=5.000000e-01
   max_dist_over_waist=8.000000e+00
   back_reject=1.000000e+00
   max_pixels_per_cell=1500
   min_pixels_per_cell=75
   I_over_U_for_match (tracking parameter)=2.000000e-01

Run from rcell::cell2

Using the same parameters.txt file

rcell2::cell2(arguments = arguments, 
              cell.command = "~/Software/cellID-linux/cell", # cell.command = system.file("cell", package = "rcell2", mustWork = T),
              dry = F, no_cores = 2)

Using parameters:
   max_dist_over_minor_axis=5.000000e-01
   max_dist_over_waist=8.000000e+00
   back_reject=1.000000e+00
   max_pixels_per_cell=1500
   min_pixels_per_cell=75
   I_over_U_for_match (tracking parameter)=2.000000e-01

Lowering background_reject_factor in parameters.txt

The lower background_reject_factor is recognized correctly by CellID.

 max_split_over_minor 0.50
 max_dist_over_waist 8.00
 max_pixels_per_cell 1500
 min_pixels_per_cell 75
 background_reject_factor 0.80 <<<<<<<<<<<<<<<<<<<<<<<<<<<<
 tracking_comparison 0.20
 align_fl_to_bf
 image_type brightfield
 bf_fl_mapping list

Using parameters:
   max_dist_over_minor_axis=5.000000e-01
   max_dist_over_waist=8.000000e+00
   back_reject=8.000000e-01 <<<<<<<<<<<<<<<<<<<<<<<<<<<<
   max_pixels_per_cell=1500
   min_pixels_per_cell=75
   I_over_U_for_match (tracking parameter)=2.000000e-01

Using comma in background_reject_factor in parameters.txt

The lower background_reject_factor with a comma as decimal separator is NOT recognized correctly by CellID.

 max_split_over_minor 0.50
 max_dist_over_waist 8.00
 max_pixels_per_cell 1500
 min_pixels_per_cell 75
 background_reject_factor 0,80 <<<<<<<<<<<<<<<<<<<<<<<<<<<<
 tracking_comparison 0.20
 align_fl_to_bf
 image_type brightfield
 bf_fl_mapping list

Using parameters:
   max_dist_over_minor_axis=5.000000e-01
   max_dist_over_waist=8.000000e+00
   back_reject=0.000000e+00 <<<<<<<<<<<<<<<<<<<<<<<<<<<<
   max_pixels_per_cell=1500
   min_pixels_per_cell=75
   I_over_U_for_match (tracking parameter)=2.000000e-01

[naikymen]

When setting:

max_dist_over_waist 6,50

CellID will read:

max_dist_over_waist=6.000000e+00

Which is effectively a rounding to integer error.

[naikymen]

Bug is confirmed, VCellID versions that use a comma as decimal separator will not pass argument values correctly to CellID if they have any decimal digits different from zero.

[gaspanico]

This could be fixed either by fixing the locale settings in VCell-ID, or by ensuring that parameter import in Cell-ID automatically replaces commas with periods. Not sure which would be better.

And a third option would be to directly "ignore" this bug, and simply have users use the Rcell2 wrapper.