Artificial Intelligence Foundation Models for Understanding Cellular Responses to Radiation Exposure
Authors:
Abraham Stroka* ([email protected]), Rebecca Weinberg, Sara Forrester, Casey Stone, Rafael Vescovi, Dan Schabacker, Thomas S. Brettin, Arvind Ramanathan, Rick Stevens (PI)
Institutions:
Argonne National Laboratory
Abstract
The use of image classification machine learning models has the potential for great impact on the speed and accuracy of medical diagnoses. The ability to accurately identify genetic perturbations based on cellular morphology would be crucial to the medical field. A large amount of research has been conducted on the effects of acute, high-dose radiation on the morphological profile of human cells. However, the effects of low-dose radiation on cellular morphology have yet to be investigated.
For the purposes of this study, researchers focused on the analysis of human umbilical vein endothelial cells (HUVEC) that have undergone low-dose radiation exposure. If phenotypic features of the HUVEC cell’s morphology can be identified using this model, it could lead to advanced and more efficient screening for low-dose radiation exposure. The team implemented a vision transformer pipeline as the image classification model for this study, specifically the mura vision transformer, which has shown reliable validation accuracy. In order to train this model, the team implemented a pipeline that utilizes the CellProfiler cell image analysis software to perform cell segmentation on HUVEC cell painting images. The CellProfiler pipeline the team developed allows researchers to stack the several channels of cell painting images and export the images of each individual cell into the vision transformer.