Presentation Time: 1:30-1:50
Home University: UNC-Chapel Hill
Research Mentor: Jason Stein, Department of Genetics
Research Title: Segmentor: Training a Machine Learning Algorithm to Optimize the Accuracy of Automatic Brain Segmentation
Individuals with autism often have increased head size, or macrocephaly. Previous studies have described periods of premature heightened brain development that are typically followed by stunts in development to be indicative of neuropsychiatric disorders. A novel and more accurate process of diagnosing macrocephaly includes counting the number of nuclei present in the brain by looking at imaging. However, it is currently unfeasible to count all of the neuron’s in a human’s brain; each person has about ~86 billion neurons, showing that it would manually take one individual 430 billion minutes to complete the segmentation. The Segmentor project involves the development of an annotation tool that will automate cell segmentation from 3D brain images to count how many nuclei are present. Currently, the goal of Segmentor is to create an open-source tool so that people can manually trace the brain images and feed the algorithm “good data.” This data will be used to train a deep learning process that will teach the program how to accurately evaluate where nuclei are without needing any additional human input. Therefore, the development and eventual use of Segmentor shows both improvement in how technology can be used for research and also for medical treatment itself. This program will be able to help improve brain segmentation techniques, which could eventually also be applied to various other parts of the body.