MYCOLab Ai Logo - gold and army green

Deep learning application to hyphae and spores identification in fungal fluorescence images

Summary

Researchers developed an artificial intelligence system that can automatically identify fungal infections in microscope images as accurately as experienced doctors. The system uses two different AI models working together to spot fungal spores, thread-like hyphae, and mycelium in fluorescence images. This technology could significantly reduce the time doctors spend analyzing samples and help ensure more accurate diagnoses, especially in hospitals with fewer experienced specialists.

Background

Fungal infections, particularly invasive fungal diseases, have increasing incidence and mortality rates worldwide. Current diagnostic methods rely on microscopy and culture but face challenges due to lack of skilled clinicians and high workloads. Deep learning has shown promise in medical image analysis, including pathological images, but fungal image analysis remains understudied.

Objective

This study aims to develop an automated detection system for fungal hyphae and spores in clinical fluorescence images using a dual-model deep learning framework combining YOLOX and MobileNet V2 models to improve diagnostic accuracy and efficiency.

Results

YOLOX achieved 85% precision and 90% recall for hyphae, and 77% precision and 85% recall for spores. MobileNet V2 achieved 83% precision and 100% recall for mycelium classification. The dual-model framework showed 92.5% precision, 99.3% recall, and 0.857 Kappa agreement with clinician evaluations.

Conclusion

The proposed dual-model framework demonstrates high accuracy in identifying fungal hyphae, spores, and mycelium in fluorescence images, achieving performance comparable to experienced clinicians. This automated system has significant clinical value for improving diagnostic accuracy and reducing clinician workload in fungal disease diagnosis.
  • Published in:Scientific Reports,
  • Study Type:Clinical Application Study,
  • Source: PMID: 40715216, DOI: 10.1038/s41598-025-11228-y
Scroll to Top