README.md

metadata

title: RBC Morphology Classifier
emoji: 🔬
colorFrom: red
colorTo: pink
sdk: gradio
sdk_version: 3.50.2
app_file: app.py
pinned: false
license: apache-2.0

🔬 RBC Morphology Classifier

AI-powered Red Blood Cell analysis for anemia detection using deep learning.

🎯 Overview

This Hugging Face Space hosts a trained EfficientNet-B0 model that classifies red blood cell (RBC) microscopy images as healthy or anemic. The model was trained on the AneRBC-II dataset containing 12,000 labeled RBC images.

📊 Model Performance

• Architecture: EfficientNet-B0 (transfer learning)
• Training Dataset: AneRBC-II (12,000 images)
  • 6,000 healthy RBC images
  • 6,000 anemic RBC images
• Validation Accuracy: 95.83%
• Test Accuracy: ~95%
• Model Size: ~16 MB
• Input Size: 224×224 RGB images
• Classes: 2 (healthy, anemic)

🚀 Usage

Web Interface

1. Upload an RBC microscopy image (JPG, PNG, or BMP)
2. Click "Classify RBC"
3. View results with confidence scores and probabilities

API Access

Python (Gradio Client):

from gradio_client import Client

client = Client("YOUR_USERNAME/rbc-morphology-classifier")
result = client.predict("path/to/rbc_image.jpg", api_name="/predict")
print(result)

Python (requests):

import requests
import base64

# Read and encode image
with open("rbc_image.jpg", "rb") as f:
    image_b64 = base64.b64encode(f.read()).decode()

# Call API
response = requests.post(
    "https://YOUR_USERNAME-rbc-morphology-classifier.hf.space/api/predict",
    json={"data": [image_b64]}
)

print(response.json())

cURL:

curl -X POST \
  https://YOUR_USERNAME-rbc-morphology-classifier.hf.space/api/predict \
  -H "Content-Type: application/json" \
  -d '{"data": ["base64_encoded_image"]}'

📈 Example Output

{
  "prediction": "anemic",
  "confidence": 0.9421,
  "probabilities": {
    "healthy": 0.0579,
    "anemic": 0.9421
  },
  "model": "EfficientNet-B0",
  "dataset": "AneRBC-II (12,000 images)",
  "validation_accuracy": "95.83%"
}

🔬 Training Details

Dataset: AneRBC-II

The model was trained on the AneRBC-II dataset, a benchmark dataset for computer-aided anemia diagnosis:

• Source: Mendeley Data
• Total Images: 12,000 (306×320 pixels each)
• Classes: Healthy (6,000) and Anemic (6,000)
• Includes: Original images, segmentation masks, CBC reports, morphology reports
• License: CC BY 4.0

Training Configuration

• Framework: PyTorch 2.1.0
• Model: EfficientNet-B0 (pretrained on ImageNet)
• Optimizer: AdamW (lr=1e-4, weight_decay=1e-4)
• Scheduler: CosineAnnealingLR (T_max=20)
• Loss: CrossEntropyLoss
• Batch Size: 32
• Epochs: 20
• Data Split: 70% train / 15% val / 15% test
• Augmentation: Random flips, rotation (20°), color jitter

Data Augmentation

Training images were augmented with:

• Random horizontal and vertical flips
• Random rotation (±20°)
• Color jitter (brightness, contrast)
• Resize to 224×224
• Normalization (ImageNet statistics)

📚 Citation

Dataset Citation

If you use this model or the AneRBC dataset in your research, please cite:

@article{shahzad2024anerbc,
  title={AneRBC dataset: a benchmark dataset for computer-aided anemia diagnosis using RBC images},
  author={Shahzad, Muhammad and Shirazi, Syed Hamad and Yaqoob, Muhammad and Khan, Zakir and Rasheed, Assad and Sheikh, Israr Ahmed and Hayat, Asad and Zhou, Huiyu},
  journal={Database},
  volume={2024},
  pages={baae120},
  year={2024},
  publisher={Oxford University Press},
  doi={10.1093/database/baae120},
  url={https://doi.org/10.1093/database/baae120}
}

Dataset Link: https://data.mendeley.com/datasets/hms3sjzt7f/1

Model Citation

@software{rbc_morphology_classifier,
  title={RBC Morphology Classifier: EfficientNet-B0 for Anemia Detection},
  author={Your Name},
  year={2024},
  url={https://huggingface.co/spaces/YOUR_USERNAME/rbc-morphology-classifier}
}

⚠️ Disclaimer

This model is intended for research and educational purposes only.

• NOT FOR CLINICAL USE: This tool should not be used as the sole basis for medical diagnosis or treatment decisions.
• Consult Professionals: Always seek the advice of qualified healthcare professionals for medical concerns.
• Accuracy Limitations: While the model achieves ~95% accuracy on test data, real-world performance may vary.
• Bias: Model performance depends on image quality, staining techniques, and population characteristics.

🔗 Related Links

• AneRBC Dataset (Mendeley): https://data.mendeley.com/datasets/hms3sjzt7f/1
• Publication (Database Journal): https://doi.org/10.1093/database/baae120
• EfficientNet Paper: https://arxiv.org/abs/1905.11946
• Timm Library: https://github.com/huggingface/pytorch-image-models

📄 License

• Model: Apache 2.0
• Dataset (AneRBC): CC BY 4.0
• Code: Apache 2.0

🤝 Acknowledgments

• Dataset Authors: Muhammad Shahzad, Syed Hamad Shirazi, Muhammad Yaqoob, et al.
• AneRBC Team: Shaukat Khanum Memorial Cancer Hospital and Research Centre
• Framework: PyTorch, Timm, Gradio, Hugging Face Spaces

📧 Contact

For questions, issues, or collaborations, please open an issue in the discussion tab or contact the Space owner.

---

Built with ❤️ for medical AI research