MohamadAbouAli/OI-PBC-Dataset

Comprehensive Benchmarking of YOLOv11 for Peripheral Blood Cell Detection

📖 Overview

This repository contains the official implementation for "Comprehensive Benchmarking of YOLOv11 Architectitectures for Scalable and Granular Peripheral Blood Cell Detection" (submitted to ArXiv). Our research provides a rigorous evaluation of YOLOv11 models for automated detection and classification of 12 peripheral blood cell types using a large-scale annotated dataset.

Key Highlights:

• ✅ Comprehensive evaluation of 5 YOLOv11 variants (Nano to XLarge)
• ✅ Large-scale dataset: 16,891 images, 298,850 annotated cells across 12 classes
• ✅ YOLOv11-Medium achieves optimal balance: mAP@0.5 of 0.934
• ✅ Publicly released dataset for advancing hematology research

🛠️ Installation & Usage

Prerequisites

Install required packages

pip install ultralytics torch torchvision opencv-python pandas numpy matplotlib

📈 Results & Recommendations

Key Findings

• YOLOv11-Medium achieves the best accuracy-efficiency trade-off
• 8:1:1 split provides better performance across all models
• Smaller models benefit more from additional training data
• Diminishing returns beyond Medium variant

Clinical Recommendation

For real-world hematology applications, we recommend:

• Model: YOLOv11-Medium
• Data Split: 8:1:1
• Performance: 93.4% mAP@0.5 with practical computational requirements

🎯 Performance Highlights

• High Precision: 92.6% precision for rare cell detection
• Excellent Recall: 93.9% recall minimizing false negatives
• Robust Performance: Consistent across all 12 cell classes
• Clinical Relevance: Suitable for integration into diagnostic workflows

Quick Access:

• 🔗 Read Paper

Citation

If you find our work or dataset useful, please consider citing our preprint:

@misc{ali2025comprehensivebenchmarkingyolov11architectures,
      title={Comprehensive Benchmarking of YOLOv11 Architectures for Scalable and Granular Peripheral Blood Cell Detection}, 
      author={Mohamad Abou Ali and Mariam Abdulfattah and Baraah Al Hussein and Fadi Dornaika and Ali Cherry and Mohamad Hajj-Hassan and Lara Hamawy},
      year={2025},
      eprint={2509.24595},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2509.24595}, 
}