Ultrastructure segmentation in electron microscopy based on energy-dispersive X-ray imaging
In brief
Element-guided AI segments multiple structures in electron microscopy without hand-drawn masks
PRISM uses element-specific maps from X-ray imaging to guide a segmentation model, producing multi-class labels in electron micrographs without manually drawn masks or image-by-image prompts. It outperformed an end-to-end model across several structures and preserved useful target information across biological and technical shifts, but its performance beyond the tested settings remains to be established.
- Journal
- Cell reports methods (Q1)
- Published
- 18 November 2026
- Study design
- Unclassified
- Evidence level
- Level 5, Expert Opinion (CEBM 5)
- Authors
- Changmin Tian, B H Peter Duinkerken, Jacob P Hoogenboom, George Azzopardi, Ben N G Giepmans, Ahmad M J Alsahaf
- PMID
- 43051160
- DOI
- 10.1016/j.crmeth.2026.101669
Why clinicians should know about it
- Picked for Histology (top studies of the week, 22 November 2026): Advanced electron microscopy segmentation for quantitative ultrastructure
Abstract
Automated segmentation of electron microscopy (EM) images is essential for quantitative analysis of biological ultrastructure. Here, we present PRISM (prediction of intracellular structures in micrographs), a two-stage method that uses element-based, structure-specific maps derived from energy-dispersive X-ray (EDX) imaging to guide EM segmentation. A generative model trained on pairs of co-registered EM images and EDX-derived structure maps learns to predict the structure maps on unseen EM images. These maps are then used to generate positive and negative prompts for SAM, resulting in multi-class segmentations without manually drawn segmentation masks or per-image manual prompting. PRISM outperforms an end-to-end model across multiple structures and retains useful target-specific information under biological and technical domain shifts. Intermediate predictions from complementary modalities therefore provide a general strategy for analysis without manual segmentation masks in multi-modal microscopy.
Abstract as published, via PubMed.
For healthcare professionals. The summary is generated by AI from the published abstract, and the evidence level is assigned automatically from the study design on the Oxford CEBM hierarchy. Neither is medical advice. Read the full paper before changing practice.