Deep learning for stenotic nares classification in brachycephalic dogs.

Brachycephalic Obstructive Airway Syndrome (BOAS) is a debilitating respiratory disorder that primarily affects brachycephalic dog breeds such as French Bulldogs and Pugs. Structural abnormalities such as everted laryngeal saccules, hyperplastic soft palate (with thickness rather than length associated with BOAS severity), and stenotic nares contribute to increased airway resistance. Stenotic nares are recognized as a major conformational risk factor associated with BOAS; however, their presence alone does not establish a clinical diagnosis, which requires comprehensive history and physical examination. Traditional veterinary assessment of nostril conformation relies largely on visual inspection, which may introduce subjectivity and inter-observer variability. This study proposes a deep learning–based approach using Convolutional Neural Networks (CNNs) for objective classification of nostril conformation in brachycephalic breeds. High-resolution nostril images from French Bulldogs and Pugs were categorized into three classes: open, mild, and severe stenosis, while acknowledging that clinical grading systems commonly include a moderate category, which was excluded due to limited representation in the dataset. Six CNN architectures—VGG16, InceptionV3, EfficientNetB0, DenseNet121, NASNetMobile, and MobileNet—were evaluated using accuracy, precision, F1-score, sensitivity, and specificity. EfficientNetB0 achieved the highest performance (96.50% accuracy), followed by VGG16 (95.67%). The findings demonstrate that deep learning can provide consistent and objective classification of stenotic nares as a key anatomical risk factor, rather than a standalone diagnostic tool for BOAS.