CellSAM model results on microscope images. Individual cells are marked in their own distinct color. Credit: Caltech
Imaging is a critical technique in biology—from identifying cancerous cells in biopsies to observing how immune cells like macrophages hunt down and destroy pathogens. Traditionally, distinguishing and labeling individual cells in images and videos has been an arduous task done by hand. Now, an interdisciplinary team of Caltech researchers has developed an artificial intelligence algorithm to identify cells in images for a wide variety of biological applications.
The new tool, called CellSAM (Cell Segment Anything Model), is the result of a collaboration between the laboratories of David Van Valen (Ph.D. ’11), an assistant professor of biology and biological engineering, Heritage Medical Research Institute Investigator, and Howard Hughes Medical Institute (HHMI) Freeman Hrabowski Scholar; and Yisong Yue, professor of computing and mathematical sciences. A paper describing the research is published in the journal Nature Methods.
“Before, students would spend countless hours identifying cells by hand or fixing an algorithm’s mistakes,” Van Valen says. “Now, our single model can do that work for you in many different applications. I’m really excited to see how this method pushes the frontier of biological discovery. There’s a lot of fascinating, interesting data that we can now collect and the previous hurdles to getting insights from those data are slowly being knocked down one at a time.”
Biological images can look vastly different, revealing phenomena such as tumor cells hidden among tissues and bacteria that secrete sticky antibiotic-resistant goo. Additionally, technological advances are ushering in an era of big data for biology.
CellSAM is the first model that can be applied to myriad different use cases, allowing researchers to identify different cell types, and to see where they are located and how they are interacting with their neighbors. Characterizing these complex dynamics is critical for understanding situations such as why a certain cancer immunotherapy may work for one person but not for another.
The CellSAM algorithm was trained on vast numbers of biological images that had been labeled by hand. The team aims to continue improving CellSAM by continuing to train it on more types of biological data. The tool is currently available for researchers to use for free.
“Approaches like CellSAM don’t just make existing image-analysis workflows more efficient—they make it possible to explore biological questions at scales that used to be impractical,” Yue says. “When you can track millions of cells across many conditions, you can start probing things like how rare cell states appear or how subtle changes in cell shape relate to treatment response. These are the kinds of insights that become accessible only when the bottlenecks in analysis are removed.”
Publication details
Markus Marks et al, CellSAM: a foundation model for cell segmentation, Nature Methods (2025). DOI: 10.1038/s41592-025-02879-w
Provided by
California Institute of Technology
Citation:
AI algorithm identifies cells across diverse biological images, cutting hours of manual labeling (2026, April 20)
retrieved 20 April 2026
from https://phys.org/news/2026-04-ai-algorithm-cells-diverse-biological.html
This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no
part may be reproduced without the written permission. The content is provided for information purposes only.