In studies involving female rodents, researchers must identify the rodents’ estrous cycle stage—an indirect measure of estrogen levels—to consider if estrogen may influence the results of their studies. When performed manually, estrous staging is a time-consuming process requiring careful examination of vaginal epithelial smears. 

A team led by Rachel Ross, M.D., Ph.D., used supervised machine learning to develop a rapid, reliable, and accessible method for monitoring the estrous cycle in female rodents. Relying on a dataset of vaginal cytology images—335 images for training purposes, 45 for validation, and 350 for testing—the researchers trained an artificial-intelligence model they call Object Detection Estrous Staging (ODES). The model proved 80% accurate on average in identifying rodent estrous stages—on par with human accuracy and previous image classification models. In addition, ODES dramatically speeds up the identification process, analyzing 100 test images in 2.67 minutes—about one-tenth of the time when done manually. It should be especially useful for large-scale neuropsychiatric studies involving female rodents. The findings were published online on January 10 in Digital Psychiatry and Neuroscience.

Dr. Ross, the paper’s corresponding author, is an assistant professor of psychiatry and behavioral sciences, of  medicine, and in the Dominick P. Purpura Department of Neuroscience at Einstein and is an attending physician at Montefiore. ODES was developed by the paper’s first authors, City College of New York undergraduate student Benjamin Babaev and Saachi Goyal, a senior at the Academy of Information Technology and Engineering in Connecticut.