Exploring the efficacy of an artificial intelligence model for fungal screening in bronchoalveolar lavage cytology specimens.

Publication/Presentation Date

1-1-2025

Abstract

INTRODUCTION: Screening for microorganisms in cytology is often limited by the rarity of organisms and the test's low sensitivity and time-consuming nature, leading to delayed results. Artificial intelligence (AI) offers potential solutions to these challenges. This study evaluates an AI model for fungal screening in digital bronchoalveolar lavage cytology specimens.

MATERIALS AND METHODS: We obtained 26 silver-stained CytoSpin slides from bronchoalveolar lavage specimens containing fungal organisms. Slides were digitized and segmented into 100 × 100 μm tiles using QuPath. Tiles were categorized into 3 classes: 925 "Fungus" tiles (hyphal and yeast forms), 527 "Debris" tiles (darkly staining acellular material), and 1121 "Background" tiles (macrophages, blood, etc.). Data augmentation generated 190 additional images per class. A pretrained VGG16 model was fine-tuned and trained over 20 epochs. Performance was assessed on a separate test set, with precision, recall, and F1-score calculated.

RESULTS: The model achieved 86% test accuracy. For the "Fungus" class, precision was 0.87, recall 0.97, and F1-score 0.92, indicating high sensitivity for fungal detection. The "Background" class achieved the highest precision (0.94), but lower recall (0.76), with an F1-score of 0.84. The "Debris" class showed a precision of 0.71, recall of 0.86, and an F1-score of 0.78.

CONCLUSIONS: Our study highlights the potential of AI-assisted fungal screening in bronchoalveolar lavage cytology, with an overall accuracy of 86%. The high recall for the "Fungus" class reflects the model's strong sensitivity across diverse fungal morphologies. Future work should focus on dataset expansion and model refinement to support integration into routine diagnostics.

Volume

14

Issue

6

First Page

419

Last Page

427

ISSN

2213-2945

Disciplines

Business Administration, Management, and Operations | Health and Medical Administration | Management Sciences and Quantitative Methods

PubMedID

40750546

Department(s)

Administration and Leadership

Document Type

Article

Link to Full Text

Share

COinS