SAM for Robust Mitochondria Instance Segmentation in Fluorescence Microscopy
Researchers have adapted the Segment Anything Model (SAM) for segmenting mitochondria in fluorescence microscopy images. The primary challenge addressed is the domain shift between natural images and microscopy data, along with a scarcity of annotated datasets. To overcome this, the team fine-tuned SAM using synthetically generated microscopy data that mimics real-world optical properties, demonstrating improved precision and dice scores on real images. AI
IMPACT Demonstrates a method for adapting foundation models to specialized scientific imaging tasks, potentially accelerating research in cell biology and related fields.