An automated root-fungus detector needed field-trained categories
WHAT THE EVIDENCE SAYS
A team testing automated root microscopy found that AMFinder struggled to distinguish complex field-stained roots. MycorrhizaFinder added classes and confidence-based human validation, improving detection of arbuscular and other mycorrhizal structures.
WHY IT MATTERS / OUR READ
Automated environmental measurement needs representative training material and human checks for underrepresented classes, not just a high aggregate accuracy.
CAVEATS / UNCERTAINTY
Accuracy varies by fungal type and training representation; high class accuracy can hide poor precision on uncommon categories. Method not yet a universal field standard.
SOURCE
Mycorrhiza — Kowal et al. (2026)
View original evidence ↗
MycorrhizaFinder addresses a laboratory-to-field transfer problem in ecological AI.
Read the Wild Lands assessment →