Methods
A fully automated computational pipeline for quantitative morphology analysis of human neural organoids from brightfield microscopy images.
Pipeline Overview
Each image passes through six sequential stages. The pipeline is deterministic and fully reproducible with a fixed random seed.
Image Acquisition
Brightfield microscopy images captured at 10.01 µm/px resolution across 10 timepoints over 30 days of culture.
Preprocessing
Gaussian smoothing and contrast normalization to reduce noise and standardize intensity across images.
Segmentation
Automatic thresholding with morphological cleanup to isolate the organoid from the background.
Feature Extraction
22 quantitative features computed from each segmented organoid covering morphology, intensity, texture, and quality.
Quality Control
Automated focus (sharpness ≥ 1.0) and signal-to-noise (≥ 1.0) checks to exclude unusable images.
Statistical Analysis
Non-parametric hypothesis testing with multiple-comparison correction for disease-vs-wildtype comparisons.
Preprocessing
Raw brightfield images undergo two normalization steps before segmentation to ensure consistent analysis across imaging conditions.
Gaussian Smoothing
σ = 2.0
Isotropic Gaussian filter reduces high-frequency noise while preserving organoid boundary features. The kernel width balances noise suppression against edge sharpness.
Contrast Stretching
2nd–98th percentile
Intensity values are rescaled to the 2nd–98th percentile range, normalizing brightness and contrast across images from different acquisition sessions and laboratories.
Segmentation
Three-step binary segmentation isolates the organoid from the background without manual annotation.
Otsu's Automatic Thresholding
Maximizes inter-class variance between foreground and background to find an optimal intensity threshold. No manual parameter tuning required.
Morphological Cleanup
Sequential morphological closing (fills small holes) and opening (removes small protrusions) operations clean up the binary mask. Both use a disk-shaped structuring element.
Largest Connected Component
Among all connected regions in the cleaned binary mask, only the largest component is retained as the final organoid segmentation. This eliminates debris and imaging artifacts.
Feature Extraction
22 quantitative features are computed from each segmented organoid, organized into four categories: morphology (9), intensity (3), texture (4), and quality control (3).
Morphology(9 features)
| Feature | Unit | Description | Biological Relevance |
|---|---|---|---|
| Area | µm² | Cross-sectional organoid area | Tissue growth and proliferation rates |
| Perimeter | µm | Outer boundary length | Boundary complexity and tissue expansion |
| Circularity | 0–1 | 4πA/P², where 1 = perfect circle | Tissue organization and self-assembly quality |
| Eccentricity | 0–1 | Ellipse eccentricity, 0 = circle | Directional growth bias and cytoskeletal polarity |
| Solidity | 0–1 | Area / convex hull area | Tissue compactness and structural integrity |
| Elongation | 0–1 | 1 − minor/major axis ratio | Cytoskeletal integrity and tissue architecture |
| Equiv. Diameter | µm | Diameter of circle with same area | Overall organoid size (shape-independent) |
| Major Axis | µm | Length of fitted ellipse major axis | Maximum growth dimension |
| Minor Axis | µm | Length of fitted ellipse minor axis | Tissue organization and cytoskeletal integrity |
Intensity(3 features)
| Feature | Unit | Description | Biological Relevance |
|---|---|---|---|
| Mean Intensity | 0–1 | Average pixel intensity in organoid | Overall optical density and tissue thickness |
| Std Intensity | 0–1 | Intensity standard deviation | Cellular differentiation heterogeneity |
| Intensity Range | 0–1 | Max − min intensity | Structural complexity and tissue maturation |
Texture(4 features)
| Feature | Unit | Description | Biological Relevance |
|---|---|---|---|
| Contrast | a.u. | Local intensity variation (Laplacian) | Local cellular architecture differences |
| Energy | a.u. | Inverse gradient variance | Tissue homogeneity and uniformity |
| Homogeneity | a.u. | Inverse Laplacian magnitude | Cellular uniformity within tissue |
| Correlation | −1 to 1 | Gradient direction correlation | Structural organization patterns |
Quality Control(3 features)
| Feature | Unit | Description |
|---|---|---|
| Passed | bool | Meets sharpness + SNR thresholds |
| Sharpness | a.u. | Laplacian variance (≥ 1.0 required) |
| SNR | ratio | Signal-to-noise ratio (≥ 1.0 required) |
Statistical Framework
Disease-vs-wildtype comparisons use non-parametric methods suitable for the moderate sample sizes typical of organoid experiments.
Hypothesis Testing
Mann-Whitney U test
Non-parametric rank-sum test comparing per-organoid aggregated feature means between disease and wildtype groups (N=16 organoids per group). Per-organoid aggregation avoids pseudoreplication from multiple images per organoid. Makes no assumption about the underlying distribution shape.
Effect Size
Rank-biserial correlation
Computed as r = 1 − 2U / (n₁ × n₂). Ranges from −1 to +1, quantifying the magnitude and direction of difference beyond statistical significance alone.
Multiple Comparisons
Bonferroni correction
Each feature is tested against 3 disease clones, yielding 3 comparisons per feature. The significance threshold is adjusted to control the family-wise error rate.
Significance Threshold
α = 0.05 after correction
Results are reported as significant only when the Bonferroni-corrected p-value falls below 0.05, ensuring robust claims that account for multiple testing.
Data Source
All analyses are performed on publicly available data from a peer-reviewed study.
Dataset
Zenodo 10301912
Schröter et al., Scientific Data (2024)
Images
1,407
brightfield microscopy
Organoids
64
individually tracked
Cell Lines
4
iPSC-derived (1 WT + 3 disease)
Timepoints
10
over 30 days
Cross-site validation: Data collected from 2 independent laboratories, enabling assessment of pipeline robustness to inter-lab variability.
Disease Models
| Clone | Gene | Disease | Biological Mechanism |
|---|---|---|---|
| wt2D | — | Wildtype Control | Normal cytoskeletal dynamics, neuronal differentiation, and tissue self-organization. |
| A1A | TUBA1A | Alpha-Tubulin Mutant | Disrupted alpha-tubulin polymerization impairs cytoskeletal dynamics, leading to irregular cell packing and altered tissue architecture. |
| B2A | TUBB2A | Beta-Tubulin Mutant | Reduced beta-tubulin function impairs microtubule dynamics, reducing proliferation and altering tissue organization. |
| TH2 | TH | Tyrosine Hydroxylase Deficient | Impaired dopamine biosynthesis disrupts neuronal differentiation, leading to compensatory proliferation with heterogeneous cellular maturation. |
Reproducibility
Every component of the analysis pipeline is fully documented and reproducible.
| Open Data | Zenodo 10301912 |
| Open Pipeline | All analysis parameters documented on this page |
| Pixel Size | 10.01 µm/px |
| Random Seed | 42 |
| Software | Python 3.14, scikit-image, OpenCV, SciPy |