# Few-shot learning for non-vitrified ice segmentation

## TL;DR

This paper introduces Ice Finder, a new tool using few-shot learning to identify crystalline ice in cryo-electron tomography, improving adaptability and efficiency.

## Contribution

The first application of meta-learning to cryo-electron tomography for ice segmentation, enabling rapid adaptation with minimal data.

## Key findings

- Ice Finder demonstrates strong domain generalization across diverse cryo-electron tomography datasets.
- The tool achieves fast processing and millisecond inference times on in situ datasets from EMPIAR.
- Few-shot learning significantly improves adaptability to new datasets with minimal examples.

## Abstract

This study introduces Ice Finder, a novel tool for quantifying crystalline ice in cryo-electron tomography, addressing a critical gap in existing methodologies. We present the first application of the meta-learning paradigm to this field, demonstrating that diverse tomographic tasks across datasets can be unified under a single meta-learning framework. By leveraging few-shot learning, our approach enhances domain generalization and adaptability to domain shifts, enabling rapid adaptation to new datasets with minimal examples. Ice Finder’s performance is evaluated on a comprehensive set of in situ datasets from EMPIAR, showcasing its ease of use, fast processing capabilities, and millisecond inference times.