Cell Detection and Segmentation Project

This project implements an algorithm for detecting and segmenting cells in microscopy images of the CAMA-1 cell line. The implementation follows a three-stage pipeline approach:

1. Foreground Mask Extraction: Separating cells from background
2. Cell Location Detection: Finding approximate locations of individual cells
3. Cell Boundary Segmentation: Segmenting individual cells using region growing

Project Structure

cell_segmentation/
├── data/
│   ├── images/            # Input RGB images (im1.jpg, im2.jpg, im3.jpg)
│   ├── gold_masks/        # Ground truth foreground masks (im*_gold_mask.txt)
│   └── gold_cells/        # Ground truth cell annotations (im*_gold_cells.txt)
├── src/
│   ├── part1_foreground.py
│   ├── part2_locations.py
│   ├── part3_boundaries.py
│   └── utils.py
├── results/
│   ├── part1/
│   ├── part2/
│   └── part3/
├── main.py
└── requirements.txt

Installation

1. Create a virtual environment (optional but recommended):

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate
   

2. Install dependencies:

   pip install -r requirements.txt
   

Usage

1. Place your data files in the appropriate directories:

   • Images in data/images/
   • Foreground masks in data/gold_masks/
   • Cell annotations in data/gold_cells/

2. Run the complete pipeline:

   python main.py
   

3. Check the results in the results/ directory:

   • Foreground masks in results/part1/
   • Cell locations in results/part2/
   • Cell segmentations in results/part3/

Algorithm Details

Part 1: ObtainForegroundMask

This part separates foreground (cells) from background using:

• Preprocessing: Gaussian blur and contrast enhancement
• Combined segmentation approach using Otsu's and adaptive thresholding
• Postprocessing with morphological operations

Part 2: FindCellLocations

This part finds approximate locations of cells using:

• Boundary detection to identify white borders between cells
• Distance transform to measure distance from each cell pixel to nearest boundary
• Regional maxima detection to identify cell centers

Part 3: FindCellBoundaries

This part segments individual cells using:

• Watershed-based region growing algorithm
• Gradient magnitude as the marking function
• Cell locations from Part 2 as initial seeds

Evaluation Metrics

• Part 1: Pixel-level precision, recall, and F1 score
• Part 2: Cell-level precision, recall, and F1 score
• Part 3: Cell-level Dice index and IoU at thresholds of 0.5, 0.75, and 0.9

Example Results

After running the pipeline, you'll find visualizations for each part in the results directory:

• Part 1: Original image, ground truth mask, and predicted mask
• Part 2: Original image, regional maxima map, and cell locations
• Part 3: Original image, ground truth segmentation, and predicted segmentation

Notes on Implementation

• This implementation uses traditional computer vision techniques without deep learning
• The algorithms can be adjusted by modifying parameters in each file
• The evaluation functions are designed to match the requirements in the assignment