SaSi: A Self-augmented and Self-interpreted Deep Learning Approach for Few-shot Cryo-ET Particle Detection

TL;DR

This work tackles the problem of few-shot 3D cryo-ET particle localization under high noise and missing wedge conditions. It introduces SaSi, a Self-augmented and Self-interpreted framework that fuses self-supervised pre-training, a Self-augmented Volume Infill strategy, and a Self-interpreted Consistency Guidance loss to enable robust particle detection with minimal labels, built on a 3D U-Net backbone and complemented by cc3d post-processing. Key contributions include adapting AugMix-inspired volume augmentation to cryo-ET, enforcing transformation-consistent segmentation without heavy ground-truth masks, and demonstrating strong improvements over state-of-the-art baselines on both simulated SHREC2021 and real cryo-ET data; SaSi achieves higher Localization F1 across multiple few-shot settings and provides a new benchmark for few-shot particle picking. The method is architecture-agnostic, compatible with CNNs and Vision Transformers, and offers practical gains in robustness and generalization for structural biology analyses with scarce annotations.

Abstract

Cryo-electron tomography (cryo-ET) has emerged as a powerful technique for imaging macromolecular complexes in their near-native states. However, the localization of 3D particles in cellular environments still presents a significant challenge due to low signal-to-noise ratios and missing wedge artifacts. Deep learning approaches have shown great potential, but they need huge amounts of data, which can be a challenge in cryo-ET scenarios where labeled data is often scarce. In this paper, we propose a novel Self-augmented and Self-interpreted (SaSi) deep learning approach towards few-shot particle detection in 3D cryo-ET images. Our method builds upon self-augmentation techniques to further boost data utilization and introduces a self-interpreted segmentation strategy for alleviating dependency on labeled data, hence improving generalization and robustness. As demonstrated by experiments conducted on both simulated and real-world cryo-ET datasets, the SaSi approach significantly outperforms existing state-of-the-art methods for particle localization. This research increases understanding of how to detect particles with very few labels in cryo-ET and thus sets a new benchmark for few-shot learning in structural biology.

Figures (3)

• Figure 1: This figure illustrates the training phase incorporating self-supervised learning using augmented pair $x', x"$, supervised learning with ground truth mask generated using weak labels, and self-interpreted using $x'$ and predicted class $m'$ from either of self-supervised and supervised learning phase.
• Figure 2: During testing and self-supervised learning, a sliding window approach with window size W and stride W/2 is applied, while particle center-based sampling is utilized for supervised learning.
• Figure 3: An illustration of the composition process of our self-augmented volume infill strategy. The input volume is filled with more self-augmented particles with different orientations and positions.