Summary

Analyze fluorescence microscopy images to automatically segment macrophages, quantify CD11b expression at the single-cell level, measure cell coverage, and count cells across experimental conditions. The workflow is optimized for irregular, amoeboid macrophage morphology and densely packed cell populations.

Input

Fluorescence microscopy images

• DAPI channel (nuclei)

• CD11b/APC channel (macrophage marker)

• Optional merged images

Output

• Single-cell segmentation masks

• Cell morphology measurements (area, eccentricity, solidity)

• Cell count per image

• Cell-covered area (% image area)

• Mean CD11b intensity per cell

• Statistical comparisons across conditions

Analysis breakdown

Dataset: BBBC020 (Broad Bioimage Benchmark Collection) — 25 fluorescence microscopy images of bone-marrow-derived macrophages across 6 time-point conditions (Kontrolle n=3, 15min n=5, 30min n=5, 1h n=5, 2h n=2, 24h n=5).
Each sample provides three registered RGB TIFF files: _c1 (CD11b/APC — signal in R channel), _c5 (DAPI — signal in B channel), and _(c1+c5) (composite). Images are 1040×1388 px, uint8 .

Channel assignment confirmed: CD11b/APC dominant in R channel (p99 = 186), DAPI dominant in B channel (p99 = 154), with no cross-talk between channels . Channel overview shown below.

1. Segmentation

Cellpose cpsam model (v4.0.1+, the successor to cyto3, pretrained on ~70K images) applied to the raw CD11b/APC R-channel (grayscale, channels=[0,0], flow_threshold=0.4, cellprob_threshold=0.0, diameter=None — auto-estimated per image). Empirical cell diameter was derived post-hoc as 2 × sqrt(mean_area / π) from each mask.

All 25 images segmented successfully; total of 1,154 cells detected across the dataset (mean 46 cells/image) .

• Area coefficient of variation ranged 0.48–0.91 across all samples — all well below the 1.5 re-segmentation trigger, confirming uniform mask quality .

• Empirical cell diameters: 116–172 px, consistent with large amoeboid macrophage morphology across conditions .

• Qualitative overlay inspection confirms accurate boundary delineation including in the dense 1h/2h fields.

  

  
  

2. Cell count

   Kontrolle: 30–36 cells/image (mean ~34); rises to 44–76 cells/image at 1h (mean ~55) and 64–71 at 2h (mean ~68); partial decline at 24h (29–49, mean ~41)

3. Area coverage

   Controls: 23–36% of image area covered

   Peaks at ~60–67% at 1h–2h; 24h range 29–65% (high variability).
   
   

   

4. CD11b mean intensity per cell

   Median per-cell CD11b intensity ranges 43–92 a.u. across all 25 images, with no strong directional trend across conditions.
   
   

   

Statistical analysis across conditions

1. Kruskal-Wallis omnibus test

• Cell count is the sole endpoint reaching statistical significance at the omnibus level (H = 13.82, p = 0.017).

• Area fraction covered (H = 7.38, p = 0.194) and mean CD11b per cell (H = 7.99, p = 0.157) are non-significant.

2. Dunn's post-hoc (Bonferroni-corrected pairwise comparisons)

No individual pairwise comparison survives Bonferroni correction across any endpoint. The lowest-p pairs for cell count are:

All remaining comparisons collapse to p = 1.000 after correction.

Cell count is the only endpoint with a statistically detectable omnibus effect across the six time points. The raw data show a clear directional trend: mean cell counts rise from ~34 (Kontrolle) to ~55–68 at 1h–2h, then partially return toward ~41 at 24h. However, this temporal pattern does not survive Bonferroni-corrected pairwise testing, which is expected given the small per-group replicate counts (n = 2–5): the conservative penalty absorbs biologically plausible effect sizes at this sample depth. This is consistent with macrophage recruitment dynamics where early time points reflect cellular influx before individual activation-state changes become detectable.

The 2h condition carries the most prominent apparent peak in both cell count (~67–71 cells per image) and area coverage (~60%); however, with only n = 2 replicates, this apparent peak is severely underpowered and must be treated as exploratory rather than confirmatory.

Area coverage and per-cell CD11b intensity show no statistically significant variation across conditions, which implies that the overall increase in CD11b signal over time is primarily a population-size effect — more cells accumulating — rather than per-cell transcriptional or translational upregulation of CD11b.

Next steps

• Increase replication at 2h (and ideally all conditions to n ≥ 5) to provide adequate power for Bonferroni-corrected pairwise testing. With current n, FDR-based correction (Benjamini-Hochberg) would be less conservative and may reveal significant pairs — consider applying it as a sensitivity analysis.

• Trend analysis: Given the clear directional increase in cell count, a non-parametric trend test (e.g., Jonckheere-Terpstra) across the ordered time points would be more statistically efficient than exhaustive pairwise comparisons.

• Per-cell CD11b: Consider testing the full per-cell distribution (using percell_all.csv, 1,154 cells) rather than per-image means, which would substantially increase statistical power for detecting intensity shifts.