SAM for Robust Mitochondria Instance Segmentation in Fluorescence Microscopy

2026-05-29

AI Summary

This work addresses the challenge of mitochondrial instance segmentation in fluorescence microscopy, which is hindered by domain shifts—such as low resolution, low contrast, and structural overlap—and the scarcity of real annotated data. For the first time, the authors fine-tune the Segment Anything Model (SAM) exclusively on fully synthetic fluorescence images, leveraging physically based microscopic image simulation combined with realistic mitochondrial morphology modeling to generate large-scale training data. Remarkably, this approach achieves robust segmentation of mitochondria in real images without any real-world annotations. Evaluated on a manually labeled test set, the method significantly outperforms strong existing baselines, demonstrating the effectiveness and novelty of a purely synthetic data–driven strategy, as validated by metrics including Dice score and precision.

Abstract

The morphological analysis of mitochondria in fluorescence microscopy (FM) is crucial for understanding cellular health, energy production, and metabolic regulation. While foundation models like the Segment Anything Model (SAM) have revolutionized natural image segmentation, their direct application to FM is hindered by a significant domain shift characterized by diffraction-limited resolution, low contrast, and complex overlapping organelle networks. Furthermore, the development of robust models is bottlenecked by a severe lack of high-quality, manually annotated instance segmentation datasets for mitochondria. In this paper, we propose a scalable solution to this data scarcity by finetuning SAM exclusively on synthetically generated FM data. We simulate realistic mitochondria data and emulate the optical properties of fluorescence microscopes to create a large-scale annotated dataset. We evaluate our fine-tuned model on a curated dataset of real, manually annotated FM images. Qualitative and quantitative analyses demonstrate that our synthetically fine-tuned model improves precision and average dice score over strong baselines. This work establishes the potential of simulation-assisted training for FM instance segmentation.

Problem

• mitochondria instance segmentation
• fluorescence microscopy
• domain shift
• data scarcity
• foundation models

Innovation

• synthetic data
• Segment Anything Model
• mitochondria segmentation
• fluorescence microscopy
• simulation-assisted training