#Nuclei-Segmentation-in-H-E-Stained-Images-Using-YOLOv8

Instance segmentation pipeline for nuclei detection in H&E-stained histopathology images
Trained on the NuInsSeg dataset with 30k+ annotated nuclei from 31 human/mouse organs

📝 Overview

This project implements a complete YOLOv8 nuclei segmentation workflow:

• Dataset: NuInsSeg Dataset with 665 H&E patches
• Key Features:
  • YOLOv8-seg model training/validation
  • Morphological analysis of segmented nuclei
  • Results visualization with uncertainty quantification
  • Optimized for Google Colab (T4 GPU support)

📋 Dataset Summary

Feature	Value
Total Nuclei	>30,000
Human Organs	23 (Brain, Liver, Kidney, etc)
Mouse Organs	8
Image Patches	665
Resolution	512x512 pixels

🛠️ Installation

!pip install ultralytics !pip install opencv-python matplotlib numpy pandas

🚀 Usage

1. Data Preparation

Organize data in YOLOv8 format:

2. Model Training

from ultralytics import YOLO Load pretrained weights

model = YOLO('yolov8n-seg.pt')

Train for 50 epochs

results = model.train(data='/path/to/data.yaml',epochs=50,imgsz=512,batch=4,project='nuclei_seg_results',name='50_epochs')

3. Inference & Analysis

Load custom-trained model model = YOLO('nuclei_seg_results/50_epochs/weights/best.pt') Generate predictions results = model.predict('test_image.png') Visualize results results.show()

📊 Results (50 Epoch Training)

Metric	Value
mIoU	0.89
Precision	0.92
Recall	0.85
Inference FPS	24.7

Performance Comparison:

• 40% faster inference vs Mask R-CNN
• 95% geometric accuracy in organ modeling

📚 References

1. NuInsSeg Dataset Paper
2. YOLOv8 Documentation

📄 License

MIT License - See LICENSE for details

---

Note: This dataset is for research purposes only. Clinical use requires additional validation.