Summary This source did not publish a separate summary. Review SKILL.md before using the skill.
Microscopy Image Analysis and Quantitative Imaging Data
Production-ready skill for analyzing microscopy-derived measurement data using pandas, numpy, scipy, statsmodels, and scikit-image. Designed for BixBench imaging questions covering colony morphometry, cell counting, fluorescence quantification, regression modeling, and statistical comparisons.
IMPORTANT : This skill handles complex multi-workflow analysis. Most implementation details have been moved to references/ for progressive disclosure. This document focuses on high-level decision-making and workflow orchestration.
When to Use This Skill
Apply when users:
Have microscopy measurement data (area, circularity, intensity, cell counts) in CSV/TSV
Ask about colony morphometry (bacterial swarming, biofilm, growth assays)
Need statistical comparisons of imaging measurements (t-test, ANOVA, Dunnett's, Mann-Whitney)
Ask about cell counting statistics (NeuN, DAPI, marker counts)
Need effect size calculations (Cohen's d) and power analysis
Want regression models (polynomial, spline) fitted to dose-response or ratio data
Ask about model comparison (R-squared, F-statistic, AIC/BIC)
Need Shapiro-Wilk normality testing on imaging data
Want confidence intervals for peak predictions from fitted models
Questions mention imaging software output (ImageJ, CellProfiler, QuPath)
Need fluorescence intensity quantification or colocalization analysis
Ask about image segmentation results (counts, areas, shapes)
BixBench Coverage : 21 questions across 4 projects (bix-18, bix-19, bix-41, bix-54)
NOT for (use other skills instead):
Phylogenetic analysis → Use tooluniverse-phylogenetics
RNA-seq differential expression → Use tooluniverse-rnaseq-deseq2
Single-cell scRNA-seq → Use tooluniverse-single-cell
Statistical regression only (no imaging context) → Use tooluniverse-statistical-modeling
Core Principles
Data-first approach - Load and inspect all CSV/TSV measurement data before analysisQuestion-driven - Parse the exact statistic, comparison, or model requestedStatistical rigor - Proper effect sizes, multiple comparison corrections, model selectionImaging-aware - Understand ImageJ/CellProfiler measurement columns (Area, Circularity, Round, Intensity)Workflow flexibility - Support both pre-quantified data (CSV) and raw image processingPrecision - Match expected answer format (integer, range, decimal places)Reproducible - Use standard Python/scipy equivalents to R functions
Required Python Packages # Core (MUST be installed)
import pandas as pd
import numpy as np
from scipy import stats
from scipy.interpolate import BSpline, make_interp_spline
import statsmodels.api as sm
from statsmodels.formula.api import ols
from statsmodels.stats.power import TTestIndPower
from patsy import dmatrix, bs, cr
# Optional (for raw image processing)
import skimage
import cv2
import tifffile
pip install pandas numpy scipy statsmodels patsy scikit-image opencv-python-headless tifffile
High-Level Workflow Decision Tree START: User question about microscopy data
│
├─ Q1: What type of data is available?
│ │
│ ├─ PRE-QUANTIFIED DATA (CSV/TSV with measurements)
│ │ └─ Workflow: Load → Parse question → Statistical analysis
│ │ Pattern: Most common BixBench pattern (bix-18, bix-19, bix-41, bix-54)
│ │ See: Section "Quantitative Data Analysis" below
│ │
│ └─ RAW IMAGES (TIFF, PNG, multi-channel)
│ └─ Workflow: Load → Segment → Measure → Analyze
│ See: references/image_processing.md
│
├─ Q2: What type of analysis is needed?
│ │
│ ├─ STATISTICAL COMPARISON
│ │ ├─ Two groups → t-test or Mann-Whitney
│ │ ├─ Multiple groups → ANOVA or Dunnett's test
│ │ ├─ Two factors → Two-way ANOVA
│ │ └─ Effect size → Cohen's d, power analysis
│ │ See: references/statistical_analysis.md
│ │
│ ├─ REGRESSION MODELING
│ │ ├─ Dose-response → Polynomial (quadratic, cubic)
│ │ ├─ Ratio optimization → Natural spline
│ │ └─ Model comparison → R-squared, F-statistic, AIC/BIC
│ │ See: references/statistical_analysis.md
│ │
│ ├─ CELL COUNTING
│ │ ├─ Fluorescence (DAPI, NeuN) → Threshold + watershed
│ │ ├─ Brightfield → Adaptive threshold
│ │ └─ High-density → CellPose or StarDist (external)
│ │ See: references/cell_counting.md
│ │
│ ├─ COLONY SEGMENTATION
│ │ ├─ Swarming assays → Otsu threshold + morphology
│ │ ├─ Biofilms → Li threshold + fill holes
│ │ └─ Growth assays → Time-lapse tracking
│ │ See: references/segmentation.md
│ │
│ └─ FLUORESCENCE QUANTIFICATION
│ ├─ Intensity measurement → regionprops
│ ├─ Colocalization → Pearson/Manders
│ └─ Multi-channel → Channel-wise quantification
│ See: references/fluorescence_analysis.md
│
└─ Q3: When to use scikit-image vs OpenCV?
├─ scikit-image: Scientific analysis, measurements, regionprops
├─ OpenCV: Fast processing, real-time, large batches
└─ Both: Often interchangeable for basic operations
See: references/image_processing.md "Library Selection Guide"
Quantitative Data Analysis Workflow
Phase 0: Question Parsing and Data Discovery CRITICAL FIRST STEP : Before writing ANY code, identify what data files are available and what the question is asking for.
import os, glob, pandas as pd
# Discover data files
data_dir = "."
csv_files = glob.glob(os.path.join(data_dir, '**', '*.csv'), recursive=True)
tsv_files = glob.glob(os.path.join(data_dir, '**', '*.tsv'), recursive=True)
img_files = glob.glob(os.path.join(data_dir, '**', '*.tif*'), recursive=True)
# Load and inspect first measurement file
if csv_files:
df = pd.read_csv(csv_files[0])
print(f"Shape: {df.shape}")
print(f"Columns: {list(df.columns)}")
print(df.head())
print(df.describe())
Area: Colony or cell area in pixels or calibrated units
Circularity: 4pi area/perimeter^2, range [0,1], 1.0 = perfect circle
Round: Roundness = 4area/(pi major_axis^2)
Genotype/Strain: Biological grouping variable
Ratio: Co-culture mixing ratio (e.g., "1:3", "5:1")
NeuN/DAPI/GFP: Cell marker counts or intensities
Phase 1: Grouped Statistics def grouped_summary(df, group_cols, measure_col):
"""Calculate summary statistics by group."""
summary = df.groupby(group_cols)[measure_col].agg(
Mean='mean',
SD='std',
Median='median',
Min='min',
Max='max',
N='count'
).reset_index()
summary['SEM'] = summary['SD'] / np.sqrt(summary['N'])
return summary
# Example: Colony morphometry by genotype
area_summary = grouped_summary(df, 'Genotype', 'Area')
circ_summary = grouped_summary(df, 'Genotype', 'Circularity')
For detailed statistical functions, see: references/statistical_analysis.md
Phase 2: Statistical Testing
Normality test needed? → Shapiro-Wilk
Two groups comparison? → t-test or Mann-Whitney
Multiple groups vs control? → Dunnett's test
Multiple groups, all comparisons? → Tukey HSD
Two factors? → Two-way ANOVA
Effect size? → Cohen's d
Sample size planning? → Power analysis
See: references/statistical_analysis.md for complete implementations
Phase 3: Regression Modeling
Polynomial (quadratic/cubic): Smooth dose-response, clear peak
Natural spline: Flexible, non-parametric, handles complex patterns
Linear: Simple relationships, checking for trends
Model comparison metrics:
R-squared: Overall fit (higher = better)
Adjusted R-squared: Penalizes complexity
F-statistic p-value: Model significance
AIC/BIC: Compare non-nested models
See: references/statistical_analysis.md for complete implementations
Raw Image Processing Workflow
When Processing Raw Images Workflow : Load → Preprocess → Segment → Measure → Export
# Quick start for cell counting
from scripts.segment_cells import count_cells_in_image
result = count_cells_in_image(
image_path="cells.tif",
channel=0, # DAPI channel
min_area=50
)
print(f"Found {result['count']} cells")
Segmentation Method Selection Cell Type Density Best Method Notes Nuclei (DAPI) Low-Medium Otsu + watershed Standard approach Nuclei (DAPI) High CellPose/StarDist Handles touching Colonies Well-separated Otsu threshold Fast, reliable Colonies Touching Watershed Edge detection Cells (phase) Any Adaptive threshold Handles uneven illumination Fluorescence Low signal Li threshold More sensitive
See: references/segmentation.md and references/cell_counting.md for detailed protocols
Library Selection: scikit-image vs OpenCV
Scientific measurements needed (area, perimeter, intensity)
regionprops for object properties
Publication-quality analysis
Easier syntax for scientists
Processing large image batches
Speed is critical
Real-time processing
Advanced computer vision features
Thresholding, filtering, morphological operations
Basic image transformations
Most segmentation tasks
See: references/image_processing.md "Library Selection Guide"
Common BixBench Patterns
Pattern 1: Colony Morphometry (bix-18) Question type : "Mean circularity of genotype with largest area?"
Data : CSV with Genotype, Area, Circularity columns
Load CSV → group by Genotype
Calculate mean Area per genotype
Identify genotype with max mean Area
Report mean Circularity for that genotype
See: references/segmentation.md "Colony Morphometry Analysis"
Pattern 2: Cell Counting Statistics (bix-19) Question type : "Cohen's d for NeuN counts between conditions?"
Data : CSV with Condition, NeuN_count, Sex, Hemisphere columns
Load CSV → filter by hemisphere/sex if needed
Split by Condition (KD vs CTRL)
Calculate Cohen's d with pooled SD
Report effect size
See: references/statistical_analysis.md "Effect Size Calculations"
Pattern 3: Multi-Group Comparison (bix-41) Question type : "Dunnett's test: How many ratios equivalent to control?"
Data : CSV with multiple co-culture ratios, Area, Circularity
Create Strain_Ratio labels
Run Dunnett's test for Area (vs control)
Run Dunnett's test for Circularity (vs control)
Count groups NOT significant in BOTH tests
See: references/statistical_analysis.md "Dunnett's Test"
Pattern 4: Regression Optimization (bix-54) Question type : "Peak frequency from natural spline model?"
Data : CSV with co-culture frequencies and Area measurements
Convert ratio strings to frequencies
Fit natural spline model (df=4)
Find peak via grid search
Report peak frequency + confidence interval
See: references/statistical_analysis.md "Regression Modeling"
Quick Reference Table Task Primary Tool Reference Load measurement CSV pandas.read_csv() This file Group statistics df.groupby().agg() This file T-test scipy.stats.ttest_ind() statistical_analysis.md ANOVA statsmodels.ols + anova_lm() statistical_analysis.md Dunnett's test scipy.stats.dunnett() statistical_analysis.md Cohen's d Custom function (pooled SD) statistical_analysis.md Power analysis statsmodels TTestIndPower statistical_analysis.md Polynomial regression statsmodels.OLS + poly features statistical_analysis.md Natural spline patsy.cr() + statsmodels.OLS statistical_analysis.md Cell segmentation skimage.filters + watershed cell_counting.md Colony segmentation skimage.filters.threshold_otsu segmentation.md Fluorescence quantification skimage.measure.regionprops fluorescence_analysis.md Colocalization Pearson/Manders fluorescence_analysis.md Image loading tifffile, skimage.io image_processing.md Batch processing scripts/batch_process.py scripts/
Example Scripts Ready-to-use scripts in scripts/ directory:
segment_cells.py - Cell/nuclei counting with watershedmeasure_fluorescence.py - Multi-channel intensity quantificationbatch_process.py - Process folders of imagescolony_morphometry.py - Measure colony area/circularitystatistical_comparison.py - Group comparison statistics # Count cells in image
python scripts/segment_cells.py cells.tif --channel 0 --min-area 50
# Batch process folder
python scripts/batch_process.py input_folder/ output.csv --analysis cell_count
Detailed Reference Guides For complete implementations and protocols:
references/statistical_analysis.md - All statistical tests, regression modelsreferences/cell_counting.md - Cell/nuclei counting protocolsreferences/segmentation.md - Colony and object segmentationreferences/fluorescence_analysis.md - Intensity quantification, colocalizationreferences/image_processing.md - Image loading, preprocessing, library selectionreferences/troubleshooting.md - Common issues and solutions
Important Notes
Matching R Statistical Functions Some BixBench questions use R for analysis. Python equivalents:
R's Dunnett test (multcomp::glht) → scipy.stats.dunnett() (scipy ≥ 1.10)R's natural spline (ns(x, df=4)) → patsy.cr(x, knots=...) with explicit quantile knotsR's t-test (t.test()) → scipy.stats.ttest_ind()R's ANOVA (aov()) → statsmodels.formula.api.ols() + sm.stats.anova_lm() See: references/statistical_analysis.md for exact parameter matching
Answer Formatting BixBench expects specific formats:
"to the nearest thousand": int(round(val, -3))
Percentages: Usually integer or 1-2 decimal places
Cohen's d: 3 decimal places
Sample sizes: Always integer (ceiling)
Ratios: String format "5:1"
Completeness Checklist Before returning your answer, verify:
Getting Help
Start with decision tree at top of this file
Check relevant reference guide for detailed protocol
Use example scripts as templates
See troubleshooting guide for common issues
All statistical implementations in statistical_analysis.md
