AI-Based System for Analysis of Electron Microscope Images in Glomerular Disease
Authors' conclusion
In this multicenter diagnostic study, TEM-AID precisely quantified glomerular ultrastructures and determined glomerular disease subtypes from TEM images, significantly enhancing diagnostic efficiency and accuracy.
Study context
- Population
- Patients with chronic kidney disease across 6 medical centers, January 2021 to December 2023 (160,727 transmission electron microscopy (TEM) images)
- Participants
- 31,670 patients160,727 TEM images
- Comparison
- TEM-AID vs clinicians unaided
- Outcome
- Segmentation performance (mean IOU, Dice), subtype classification accuracy, AUC and human-AI diagnostic concordance
Methodological detailsmulticenter, external testing, human-AI test set
- Multicenter
- External testing
- human-AI test set
Performance
As reported in the abstract
- Segmentation mean IOU, mean (SD)
- 0.835 (0.062)
- Segmentation Dice, mean (SD)
- 0.874 (0.023)
- Subtype classification accuracy, internal validation
- 0.911 (0.904-0.918) (95% CI)
- Subtype classification accuracy, external tests
- 0.895 to 0.914
- Macro-AUC across cohorts
- 0.972 to 0.989
Further findings
| Outcome | Values |
|---|---|
| Human-AI test set (7 glomerular disease subtypes) | TEM-AID accuracy 0.886 (95% CI 0.859–0.912) · AUC 0.963 [95% CI, 0.937-0.989] |
- Human-AI test set (7 glomerular disease subtypes) — Clinicians' accuracy improvement when using TEM-AID, mean (SD)
- 11.7% (5.2%)
- Human-AI test set (7 glomerular disease subtypes): 454 patients
Human-AI test set (7 glomerular disease subtypes)
- TEM-AID accuracy exceeded clinicians' unaided performance
Original publication
AI-Based System for Analysis of Electron Microscope Images in Glomerular Disease. JAMA Netw Open. Published 1 October 2025. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →