Fraiwan Mohammad, Mukbel Rami, Kanaan Dania
Department of Computer Engineering, Jordan University of Science and Technology, Irbid, Jordan.
College of Veterinary Medicine, Jordan University of Science and Technology, Irbid, Jordan.
PLoS One. 2025 Apr 3;20(4):e0320224. doi: 10.1371/journal.pone.0320224. eCollection 2025.
Sandflies are vectors for several tropical diseases such as leishmaniasis, bartonellosis, and sandfly fever. Moreover, sandflies exhibit species-specificity in transmitting particular pathogen species, with females being responsible for disease transmission. Thus, effective classification of sandfly species and the corresponding sex identification are important for disease surveillance and control, managing breeding/populations, research and development, and conducting epidemiological studies. This is typically performed manually by observing internal morphological features, which maybe an error-prone tedious process. In this work, we developed a deep learning artificial intelligence system to determine the gender and to differentiate between three species of two sandfly subgenera (i.e., Phlebotomus alexandri, Phlebotomus papatasi, and Phlebotomus sergenti). Using locally field-caught and prepared samples over a period of two years, and based on convolutional neural networks, transfer learning, and early fusion of genital and pharynx images, we achieved exceptional classification accuracy (greater than 95%) across multiple performance metrics and using a wide range of pre-trained convolutional neural network models. This study not only contributes to the field of medical entomology by providing an automated and accurate solution for sandfly gender identification and taxonomy, but also establishes a framework for leveraging deep learning techniques in similar vector-borne disease research and control efforts.
白蛉是几种热带疾病的传播媒介，如利什曼病、巴尔通体病和白蛉热。此外，白蛉在传播特定病原体物种方面表现出物种特异性，雌性负责疾病传播。因此，对白蛉物种进行有效分类并进行相应的性别鉴定，对于疾病监测与控制、管理繁殖/种群、研究与开发以及开展流行病学研究都很重要。这通常是通过观察内部形态特征手动进行的，这可能是一个容易出错的繁琐过程。在这项工作中，我们开发了一个深度学习人工智能系统，用于确定性别并区分两种白蛉亚属的三个物种（即亚历山大白蛉、巴氏白蛉和塞尔吉白蛉）。我们使用了两年时间里在当地野外捕获并制备的样本，基于卷积神经网络、迁移学习以及生殖器和咽部图像的早期融合，我们在多种性能指标上以及使用广泛的预训练卷积神经网络模型时都实现了卓越的分类准确率（超过95%）。这项研究不仅通过为白蛉性别鉴定和分类学提供一种自动化且准确的解决方案，为医学昆虫学领域做出了贡献，还为在类似的媒介传播疾病研究和控制工作中利用深度学习技术建立了一个框架。
