Added feature to phytoMorph that identifies initial root at the start of a time-lapse
This program will use a Machine Learning (ML) algorithm to segment a plant root in the first frame of a time-lapse experiment. It is intended to be an additional feature for the Spalding lab's PhytoMorph program, which performs a kinematics analysis on growing roots.
The current system requires the user to manually click along the midline of a root at the beginning of a time-lapse experiment (frame 1) in order to initialize PhytoMorph's segmentation algorithm that forms a contour around the user's midline. The manual aspect of PhytoMorph hinders it from being truly high-throughput and would make a large screen of many genotypes difficult to analyze.
SoulSearch is intended to address this issue by using an ML algorithm to perform the initial segmentation step typically performed by a human. Currently (as of 9/19/2018), I am strategizing multiple methods in which this ML algorithm will function: 1) Imitate the clicks a user would normally make 2) Manually draw contours around root images 3) Form a contour from a user-defined midline (similar to Method 1)
I am currently in favor of Method 1, since there is already an algorithm in PhytoMorph that will generate a contour around a user's clicks. This means I wouldn't have to draw a ton of contours around roots.
Clone this repository into your desired location with
git clone https://github.com/jbustamante35/soulsearch
-
Import minimal suite of functions from rootKinematics to get pipeline running as expected
a) Refactor the pipeline to minimize overhead and compute power
b) Write headless version to run from commandline -
Extract clicking coordinates from finished datasets
a) Midlines
b) Quiecent Centers -
Write a learning algorithm that predicts midlines and qc from root images
Julian Bustamante, Cellular and Molecular Biology ([email protected]
University of Wisconsin - Madison
Department of Botany