Final Phase

 Hello all!

The Google Summer of Code 2020 is now near its completion and it is time to review all the work that I have done these three months:

• Created a dataset of individual, microscopic cell images and corresponding ground truth images from the base dataset consisting of wide-field epifluorescent images of mouse neuroblastoma cells (cultured neurons) with cytoplasmic (phalloidin) stain and a set of manual segmentations.
• Implemented the deep learning model for biomedical image segmentation, U-Net, using the Deeplearning4j library.
• Created a Command Line Interface version for XitoSBML. 

In short, this project implements a system that automatically generates spatial SBML models by comprehensively segmenting microscopic images of cells using the deep learning algorithm U-Net in combination with XitoSBML.

Project repositories:

• https://github.com/Medha-B/unetdl4j
• https://github.com/Medha-B/XitoSBML

Commit List:

• https://github.com/Medha-B/unetdl4j/commits/master?author=Medha-B

• https://github.com/Medha-B/XitoSBML/commits/master?author=Medha-B

This week, the most important task was to complete the code documentation for both the unetdl4j repository (https://github.com/Medha-B/unetdl4j) and the XitoSBML repository (https://github.com/Medha-B/XitoSBML). For this, I added Javadocs for these repositories. Please see https://github.com/Medha-B/unetdl4j/tree/master/doc and https://github.com/Medha-B/XitoSBML/tree/master/apidocs for final code documentation.

Additionally, I updated the README.md for the unetdl4j repository at https://github.com/Medha-B/unetdl4j/blob/master/README.md.  

Important links:

• The dataset comprised of 300 cellular images and corresponding ground truth images: https://drive.google.com/drive/folders/1UUq6W-3P7Mg-eSE6_UJSCQaC8Xazc3zH
• The dataset comprised of 100 cellular images and corresponding ground truth images:https://drive.google.com/drive/folders/1u3SgJYb1LObpboEKkURQr3Mh7FrPrf_8
• The dataset comprised of 44 cellular images and corresponding ground truth images:https://drive.google.com/drive/folders/1Ox0fi1V9dwBXPHisgLc9kjaIfZFZ27dy
• The dataset comprised of 20 cellular images for testing: https://drive.google.com/drive/folders/1lNphWDWUDq6U4K25kP-zHL8U4ETawuDE
• The dataset comprised of 20 inferred images (for a model trained with 300 images over 100 epochs) along with ground truth: https://drive.google.com/drive/folders/1mPoAuW98VMYrqDFtz7QdX9ZA43iuUylU
• The dataset comprised of 20 inferred images (for a model trained with 100 images over 50 epochs) along with ground truth: https://drive.google.com/drive/folders/1ZetKvrxNPULv_AejnTvnxEi0zlk2Zg4B
• The saved weights for models trained for 300 images over 100 epochs and for 100 images over 50 epochs: https://drive.google.com/drive/folders/1-uX5HFjyW-e5-NnNUowwr5LiOEVh91YE
• The values for Intersection over Union and Dice Coefficient for images inferred from the trained model: https://drive.google.com/file/d/1B3yqWxzfooUaqsUNn3q5lgtyfJgJNXN-/view

Post GSoC activities:

• Closing unresolved issues

1. https://github.com/Medha-B/XitoSBML/issues/14

2. https://github.com/Medha-B/XitoSBML/issues/15

• Completing CrossVal.java

The implementation of CrossVal.java is currently incomplete and I will focus on completing this as a part of the post-GSoC activity.

Participating in Google Summer of Code has been an amazing learning experience. I had a lot of fun experimenting with different ways of writing code, learning from beautifully written programs. Who had thought that fixing bugs and resolving issues could be so engaging? The project milestones which had looked impossibly daunting earlier became easier to achieve as I became more proficient through the weeks. 

I thank my mentors, Dr. Akira Funahashi, Mr. Yuta Tokuoka, and Mr. Kaito Ii for their extreme kindness and enthusiasm throughout. It is truly because of their help and support that this project reached its conclusion. 

Cheers to the most memorable experience...

Signing off...

Ta-Ta!