Inspiration

Neuroscience is a field that is being augmented by recent work in deep learning. This project presents one such way in which traditional neuroscience research can be assisted by deep learning technologies.

What it does

We built a model that augments microscopic images of neuron clusters to improve downstream accuracy of an image segmentation model which identifies neurons from their surroundings.

How we built it

We used Python and Keras in addition to outside data science and computer vision libraries like Numpy and PIL to explore, transform, and augment our data

Challenges we ran into

There were lots of issues with data cleanliness that we had to resolve. An interesting challenge related to visualizing the data involved processing it such that the inputs and outputs of the model could actually be interpreted visually. Understanding these differences was insightful in data augmentation and modifying inputs.

Accomplishments that we're proud of

We managed to marginally improve accuracy.

What we learned

We learned how to use computer vision in combination with deep learning.

What's next for SnekNet Brain Segmentation

Flood-filling or other pre-processing

Built With

Share this project:

Updates