The CLEXM consortium is proud to share that one of our students Shao Sen (Vincent) Chueh, has a preprint of his research available. Here’s what Vincent had to say about it:
“I’m glad to share that I’ve drafted my recent research into a new preprint “SXTractor: A Self-Supervised Feature Extractor of Soft X-Ray Images That Enables Few-Shot Tomogram Segmentation”
Soft X-ray tomography (SXT) is a powerful imaging modality that bridges the resolution gap between light and electron microscopy by visualizing whole-cell structures in near-native states. However, the limited availability of annotated data has slowed down the development of deep learning methods for SXT analysis.
In this work, we present SXTractor, a self-supervised feature extractor that establishes a robust foundation for downstream SXT image processing tasks. One of its most exciting applications is few-shot tomogram segmentation:
1) With only 5–10 expert-labeled slices, SXTractor enables accurate segmentation of entire tomograms, drastically reducing manual annotation effort.
2) Different from organelle-specific models, which often struggle to generalize across datasets with limited annotated training data, few-shot segmentation offers a flexible and generalizable strategy — allowing biologists to adapt models to new data with minimal additional labeling.
3) Beyond segmentation, SXTractor serves as a non-task-specific encoder, making it adaptable to a wide range of SXT image processing tasks.
I believe this approach represents a shift toward more generalizable and data-efficient deep learning in SXT, helping unlock the full potential of this emerging imaging modality for biological discovery.
I want to extend my sincere gratitute to my supervisors Sergey Kapishnikov and Jeremy Simpson for all the help and support, and also Madeleen Brink for helping with the samples used in this research.”