DropDrop

Automated pipeline for detecting droplets and inclusions (beads) in microscopy z-stacks using Cellpose segmentation and morphological analysis.

Tailored for the EVOS M5000 Imaging System.

Installation

# Using uv (recommended)
uv pip install dropdrop

# Using pip
pip install dropdrop

GPU support (CUDA)

On Linux/Windows with an NVIDIA GPU, install CUDA-enabled PyTorch before installing DropDrop:

# Install PyTorch with CUDA 12.6 (adjust version for your driver)
uv pip install torch torchvision --index-url https://download.pytorch.org/whl/cu126

# Then install DropDrop
uv pip install dropdrop

On macOS, GPU acceleration via Metal (MPS) is included in the default PyTorch build — no extra steps needed.

From source

git clone https://github.com/yourusername/dropdrop.git
cd dropdrop
uv pip install -e .

Quick Start

# Single directory — interactive prompts for settings
dropdrop ./images

# Process only first 5 frames (for testing)
dropdrop ./images -n 5

# With interactive editor
dropdrop ./images -e

# Multiplex mode — batch process subdirectories
dropdrop -m ./samples

Usage

Single Mode

# Basic run (prompts for settings interactively)
dropdrop ./images

# Custom output directory
dropdrop ./images ./results/my_project

# With editor and archive
dropdrop ./images -e -z

Multiplex Mode

Process multiple sample directories at once. Each subdirectory is labeled interactively and processed as a separate sample, then combined into a multiplexed report.

dropdrop -m ./samples_parent_dir
dropdrop -m ./samples_parent_dir -z    # Archive result

Resume (Resurrect)

If a multiplex run is interrupted, resume from where it left off:

dropdrop -r

Cache Control

dropdrop ./images --no-cache       # Disable caching
dropdrop ./images --clear-cache    # Clear cache before run

Interactive Editor

The editor (-e) allows manual correction of detected inclusions:

Disabled droplets (gray with X) are excluded from the final results.

Output Structure

Single mode

results/<YYYYMMDD>_<label>/
  data.csv                  # Raw detection data
  summary.txt               # Settings and statistics
  report.png                # Combined report with sample frames
  size_distribution.png     # Droplet diameter histogram
  poisson_comparison.png    # Bead distribution vs theoretical

Multiplex mode

results/<YYYYMMDD>_multiplex/
  data.csv                  # Merged data with sample column
  summary.txt               # Per-sample statistics
  summary_report.png        # Comparison table, overlaid plots, sample collage
  size_distribution.png     # Overlaid diameter histograms
  poisson_comparison.png    # Overlaid inclusion distributions

data.csv columns

Architecture

CLI
  -> Detection (per sample) -> .tmp_<label>/data.csv + sample_*.png
  -> Analysis.run(output_dir)  -- auto-discovers .tmp_* dirs
       1 sample  -> single report
       2+ samples -> multiplex report
  -> Cleanup .tmp_* dirs
  -> Archive (optional)

Configuration

Create config.json in your working directory to customize detection parameters:

{
  "cellpose_flow_threshold": 0.4,
  "cellpose_cellprob_threshold": 0.0,
  "erosion_pixels": 5,
  "kernel_size": 7,
  "tophat_threshold": 30,
  "min_inclusion_area": 7,
  "max_inclusion_area": 50,
  "edge_buffer": 5,
  "min_droplet_diameter": 80,
  "max_droplet_diameter": 200,
  "px_to_um": 1.14,
  "cache": {
    "enabled": true,
    "max_frames": 100
  }
}

Parameters

Requirements

• Python 3.12+
• CUDA-capable GPU (recommended for Cellpose)

License

MIT