๐ค AI Summary
Cryo-electron tomography (CryoET) particle picking heavily relies on labor-intensive manual annotation, leaving vast amounts of unlabeled data underutilized and limiting scalability. To address this label scarcity, we propose a label-efficient semi-supervised learning framework. First, we design an end-to-end heatmap-based detection model trained under keypoint detection supervision for precise subtomogram localization. Second, we introduce a teacherโstudent collaborative training scheme augmented with multi-view pseudo-label generation to enforce geometric and consistency constraints across tilted views. Third, we develop a CryoET-specific DropBlock augmentation tailored to the anisotropic noise and missing-wedge artifacts inherent in tomographic data. Evaluated under extreme label sparsity (e.g., only 1% annotated samples), our method achieves a 10% F1-score improvement over fully supervised baselines on the large-scale CZII dataset, significantly enhancing both unlabeled data utilization and automated particle picking efficiency.
๐ Abstract
Cryogenic Electron Tomography (CryoET) combined with sub-volume averaging (SVA) is the only imaging modality capable of resolving protein structures inside cells at molecular resolution. Particle picking, the task of localizing and classifying target proteins in 3D CryoET volumes, remains the main bottleneck. Due to the reliance on time-consuming manual labels, the vast reserve of unlabeled tomograms remains underutilized. In this work, we present a fast, label-efficient semi-supervised framework that exploits this untapped data. Our framework consists of two components: (i) an end-to-end heatmap-supervised detection model inspired by keypoint detection, and (ii) a teacher-student co-training mechanism that enhances performance under sparse labeling conditions. Furthermore, we introduce multi-view pseudo-labeling and a CryoET-specific DropBlock augmentation strategy to further boost performance. Extensive evaluations on the large-scale CZII dataset show that our approach improves F1 by 10% over supervised baselines, underscoring the promise of semi-supervised learning for leveraging unlabeled CryoET data.