Back to search

Boston University

Development and evaluation of a deep learning artificial intelligence pipeline for onychomycosis detection in whole slide images: a proof-of-concept study

Abstract

dc:description.abstract

BACKGROUND: Onychomycosis is a fungal infection of the nail unit that disproportionately affects older adults and individuals with comorbidities such as diabetes and peripheral vascular disease. Despite its high prevalence, histopathological diagnosis of onychomycosis remains a time-intensive task, often challenged by subtle morphology and staining artifacts. While artificial intelligence (AI) has transformed cancer diagnostics in digital pathology, its applications in fungal detection remain limited. This study presents a proof-of-concept exploration into the development of an AI-based image analysis model for detecting fungal elements in Periodic acid-Schiff-Alcian Blue (PAS-AB) stained whole slide images (WSIs) of nail biopsy specimens, aiming to improve diagnostic efficiency and accuracy in dermatopathology. METHODS: A total of 100 PAS-AB stained nail biopsy slides were scanned into WSIs, of which 74 met inclusion criteria after quality screening. From this dataset, ten high-quality slides were selected for model training based on defined thresholds for fungal clarity, staining contrast, and overall tissue quality. The AI model was developed on Aiforia, a cloud-based digital pathology platform, using manual annotations of both fungal and non-fungal structures to teach the algorithm meaningful morphological distinctions. The model’s training process involved multiple iterations of expert-guided annotation, retraining, and verification to improve the model’s ability to distinguish fungal elements from background artifacts. Model performance was evaluated through a blind batch analysis of 57 independent WSIs. RESULTS: The developed model demonstrated strong sensitivity (81.43%) and high tissue detection accuracy (>99%), though it exhibited moderate precision (53.97%) in classifying fungal elements. False positives were primarily associated with boundary regions and staining gradients, while false negatives arose from faint or degraded fungal morphologies not well-represented in the training set. Retraining with true negative slides significantly reduced false positive rates in negative samples. These findings underscore the feasibility of applying AI to fungal detection and identify key areas for refinement, including annotation variability, gain threshold tuning, and expansion of fungal morphological diversity in training datasets. CONCLUSION: This study establishes a foundational AI pipeline for detecting onychomycosis in digital pathology slides. While not yet optimized for clinical deployment, the model’s strong sensitivity and spatial accuracy support its potential role as an assistive screening tool. Limitations in fungal precision highlight the need for more diverse training inputs and careful threshold calibration. As the first AI fungal detection model developed on a major deep learning platform for digital pathology, this work contributes to an emerging field in digital dermatopathology and paves the frontier for future studies aiming to integrate AI into fungal diagnostics.

Degree

thesis:*
Name thesis:degree_name
Master of Science
Level thesis:degree_level
masters
Discipline thesis:degree_discipline
Medical Sciences
Grantor
Boston University
Year dc:date.issued
2025

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Hua, Carolyn
Advisors dc:contributor.advisor
  • Mellott, Tiffany
  • Bach, Dylan

Subjects

dc:subject × 8

Rights

Language dc:language.iso
en_US

Identifiers

dc:identifier.*
Handle dc:identifier.uri
https://hdl.handle.net/2144/53161

Chain of custody

source
Harvested from
Boston University
Base URL
open.bu.edu/oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Hua, Carolyn. Development and evaluation of a deep learning artificial intelligence pipeline for onychomycosis detection in whole slide images: a proof-of-concept study. masters thesis, Boston University, 2025. https://hdl.handle.net/2144/53161