Inspiration
We are all computational scientists working on applying quantitative methods to biology. We thought the filament detection challege was the one that most closely matched our interests and skills.
What it does
It detects the formation of a protein filament in cells from microscopy images.
How we built it
We manually labelled our dataset of microscopy images. We ran a toy simulation to obtain synthetic data as a test set for evaluation. We tried different methods to segment microscopy images. We evaluate them with the same metric based on percentage overlap of pixels between manual masks and predicted masks.
Challenges we ran into
Curating dataset for testing. Coming up with a toy-model to simulate synthetic microscopic images. Designing a metric to evaluate segmentation performance.
Accomplishments that we're proud of
Having running code to simulate data so that it can be used to test our segmentation methods. Having found a method (Ilastik) that gives us confidence of filament detection. Having benchmarked different methodologies against Ilastik.
What we learned
We learnt what are some current approaches to segmentation. We learnt that we should have started integrating the work everyone has done earlier. We also had never worked together or done a hackathon before so it was a steep learning curve to manage to work together, especially considering we have all worked on similar tasks.
What's next for Filament detection
We should add the rest of the methods we have tested to the Web Server so they can be compared as well. We can redefine the evaluation metric for detection and for calculating the length of the filament and for the time it persists throughout the data collection.
Log in or sign up for Devpost to join the conversation.