OpenFungi: A Machine Learning Dataset for Fungal Image Recognition Tasks.

Area of Science:

• Microbiology
• Computer Science
• Machine Learning

Background:

• Advancements in medical machine learning rely on accessible datasets.
• Many prior studies lack reproducibility due to closed or proprietary data.
• A gap exists in open, high-quality datasets for microbiology image recognition.

Purpose of the Study:

• To introduce OpenFungi, an open database of fungal images.
• To facilitate reproducible research in fungal identification and classification.
• To demonstrate the utility of the dataset for machine learning applications.

Main Methods:

• Cultivated fungal genera from food products (spices, cereals).
• Collected high-quality macroscopic and microscopic images.
• Utilized a convolutional neural network for image classification.
• Performed experimental analysis with six performance metrics across three validation scenarios.

Main Results:

• The OpenFungi dataset enabled high accuracy in fungal classification.
• Macroscopic dataset classification achieved 99.79% accuracy, 99.63% F1 score.
• Microscopic dataset classification achieved 97.82% accuracy, 95.20% F1 score.
• The model demonstrated robustness and generalization on smaller datasets.

Conclusions:

• OpenFungi addresses the need for reproducible research in fungal image analysis.
• The dataset supports the development of accurate machine learning models for microbiology.
• The findings highlight the potential of open datasets in advancing medical AI.