Instance Segmentation (Microscopy)

Overview

Developed instance segmentation models to detect densely packed cells in 3D fluorescent microscopy images.

Responsibilities

• Processed multi-channel microscopy data (DAPI channel)
• Implemented DCAN / Mask R-CNN models
• Handled high-density overlapping instances

Approach

• Contour-aware segmentation (DCAN)
• Instance segmentation (Mask R-CNN)
• 3D-aware preprocessing

Pipeline

flowchart LR
    A[Microscopy Images] --> B[Channel Extraction]
    B --> C[Preprocessing]
    C --> D[Model Training]
    D --> E[Instance Segmentation]

Tech

TensorFlow · Mask R-CNN · DCAN

Impact

• Improved segmentation accuracy in dense cell environments
• Enabled quantitative biological analysis
• Reduced manual labeling effort

Sample Result