da Silva Andrade Gustavo, Hirokawa Higa Gabriel Toshio, da Silva Ribeiro Jarbas Felipe, Medeiros Ramos Carvalho Joyce Katiuccia, Gonçalves Wesley Nunes, Naka Marco Hiroshi, Pistori Hemerson
Universidade Federal de Mato Grosso do Sul, Campo Grande, Brazil.
Universidade Católica Dom Bosco, Campo Grande, Brazil.
Sci Rep. 2025 Mar 21;15(1):9800. doi: 10.1038/s41598-025-92769-0.
Nasal stenosis in bulldogs significantly impacts their quality of life, making early diagnosis crucial for effective treatment. This study developed an automated deep learning model to classify the severity of nasal stenosis using 1020 images of bulldog nostrils, including both real and AI-generated samples. Five neural network architectures were tested across three experiments, with DenseNet201 achieving the highest median F-score of 54.04%. The model's performance was directly compared to trained human evaluators specializing in veterinary anatomy, achieving comparable levels of accuracy and reliability. These results demonstrate the potential of advanced neural networks to match human-level performance in diagnosis, paving the way for enhanced treatment planning and overall animal welfare.
斗牛犬的鼻腔狭窄严重影响它们的生活质量，因此早期诊断对于有效治疗至关重要。本研究开发了一种自动化深度学习模型，使用1020张斗牛犬鼻孔图像（包括真实样本和人工智能生成的样本）对鼻腔狭窄的严重程度进行分类。在三个实验中测试了五种神经网络架构，其中DenseNet201的中位数F分数最高，为54.04%。该模型的性能与专门从事兽医解剖学的训练有素的人类评估者进行了直接比较，达到了相当的准确性和可靠性水平。这些结果证明了先进神经网络在诊断中达到人类水平性能的潜力，为改进治疗方案和提高动物整体福利铺平了道路。
