Determination of Fungiform Papilla Number Using Deep Learning Methods

Area of Science:

• Biomedical Imaging
• Artificial Intelligence in Medicine
• Computational Biology

Background:

• Fungiform papillae (FP) density and morphology are potential biomarkers for taste function and systemic diseases.
• Manual counting of FP is subjective and time-consuming, hindering large-scale research and clinical applications.

Purpose of the Study:

• To develop and validate a deep learning-based method for automatic detection and counting of fungiform papillae (FP).
• To provide an objective, reproducible, and efficient AI tool for quantitative FP assessment.

Main Methods:

• A deep learning object detection model (Ultralytics YOLOv11) was trained on 177 annotated tongue images.
• Transfer learning, nested cross-validation, and early stopping were employed for model optimization.
• Performance was evaluated using precision, recall, F1 score, mean absolute error, and root mean square error on an independent test set.

Main Results:

• The YOLOv11 model achieved a balanced detection performance with 0.678 precision, 0.740 recall, and 0.707 F1 score.
• The model demonstrated reliable counting accuracy with a mean absolute error of 37.52 and root mean square error of 43.83.
• The developed AI method showed improved generalization and robustness compared to existing studies.

Conclusions:

• The proposed YOLOv11-based deep learning model offers a fast, accurate, and reproducible alternative to manual fungiform papillae counting.
• This AI approach has the potential to support large-scale clinical and research applications utilizing FP analysis as a health status biomarker.