Skip to content

pOSNode/MSc-Project

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2 Commits
 
 
 
 

Repository files navigation

🧠 R–Net: Regression-Based CNN for Microscopy Image Cell Quantification

Quantification

Dissertation PDF

R–Net is a custom-built Convolutional Neural Network (CNN) developed for automated, segmentation-free quantification of cell types in microscopy images. Created as part of an MSc Biomedical Engineering project at Imperial College London, R–Net achieves high predictive accuracy while significantly reducing the need for manual processing.


🚀 Highlights

  • 🧬 Segmentation-Free Approach
    Directly predicts percentage composition of 3 cell types from raw microscopy images.

  • 🧠 Custom CNN (R–Net)
    Multi-block architecture inspired by human vision hierarchy, using stacked convolutions, batch norm, and dropout.

  • 📊 Superior Accuracy
    R² = 0.7576, MSE = 0.0072, outperforming VGG16, ResNet, MobileNet, and a baseline MNIST-derived model.

  • ⚙️ End-to-End Workflow
    Includes synthetic dataset generation, data augmentation, model training, evaluation, and visualisation.


📄 Dissertation

The full methodology, results, and discussion are available in the final MSc dissertation:

📥 Download PDF
📅 Submitted: September 2023
🏫 Institution: Imperial College London
👨‍🏫 Supervisors: Prof. Anil Bharath, Dr. Faraz Janan


🧪 Performance Summary

Model MSE MAE MAPE RMSE Time (hrs)
R–Net 0.0072 0.0684 0.7576 67.26% 0.0849 ~1.5
ResNet 0.0107 0.0899 0.4025 ~4
VGG16 0.0116 0.0920 0.3215 ~4
MobileNet 0.0278 0.1017 0.0959 ~3
MNIST Baseline 0.0316 0.1051 0.3921 112.6% 0.1778 ~1.5

🧰 Tools & Technologies

  • Python · TensorFlow · Keras · OpenCV · ImgAug · scikit-learn
  • Jupyter Notebook · Matplotlib

🔧 Future Directions

  • Add skip connections (e.g., ResNet-style)
  • Apply to real microscopy data for robustness testing
  • Tune regularisation (L1/L2), augmentations, and learning rates

📜 Citation

If referencing this project or methodology:

Siddiqi, O. (2023). R–Net: A Novel Approach Towards the Cell Quantification of Microscopy Images Using a Regression-based Network. MSc Dissertation, Imperial College London.


🙏 Acknowledgements

  • Prof. Anil Bharath – Dataset generation & academic supervision
  • Dr. Faraz Janan – Research guidance and technical feedback

© 2023 Owais M Siddiqi — owaissiddiqi.co.ukLinkedInomsiddiqi01@gmail.com

About

R–Net is a custom-built Convolutional Neural Network (CNN) developed for automated, segmentation-free quantification of cell types in microscopy images. Created as part of an MSc Biomedical Engineering project at Imperial College London, R–Net achieves high predictive accuracy while significantly reducing the need for manual processing.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors