Reaching Out: Identifying Filopodia in Chick Neural Crest Cells with Machine Learning
Authors: Benjamin Lee, Gregory Arruiza, Taylor Kan
Faculty Supervisor: Ilmi Yoon
Department: Computer Science
Communication is vital to the 'communities' of cells in multicellular organisms, helping regulate proliferation, specification, and differentiation. To coordinate processes such as development or wound healing, cells communicate through various means, ranging from diffusion, cell-cell contact, secretion, to more recently filopodia. Filopodia are long-range actin protrusions elongating from producing to receiving cells, which can travel anywhere between 5 to 35 microns. Despite this, the role of filopodia in cell signaling remains poorly understood, but research from other labs has suggested a correlation between filopodia density and signaling activity. However, current methods of quantifying filopodia are inefficient, as these protrusions need to be manually counted. This project seeks to leverage advancements in deep learning to automate this process. To achieve this, we will build and test various models including Neural Networks (NN), Convolutional Neural Networks (CNN), as well as other pre-trained models to recognize filopodia in chick neural crest cells. Algorithms will be evaluated through accuracy, Jaccard, and Dice Coefficient. The findings of this project will help us understand the relationship between filopodia density and cell signaling enabling us to automate the quantification process.