Rabies remains a major public health concern across many African and Asian countries, where accurate diagnosis is essential for effective epidemiological surveillance. The gold‑standard diagnostic methods rely on fluorescence microscopy, a technique that demands skilled laboratory personnel for accurate interpretation. Such expertise is often scarce in regions with low annual sample volumes.
To address these challenges, the authors present an automated, AI‑driven diagnostic system. The pipeline performs fluorescent image analysis through transfer learning with four deep‑learning architectures: EfficientNetB0, EfficientNetB2, VGG16, and Vision Transformer (ViTB16).
The study evaluates three distinct data‑augmentation strategies on a dataset of 155 microscopic images (123 positive and 32 negative). Results indicate that TrivialAugmentWide is the most effective augmentation technique, preserving critical fluorescent patterns while improving model robustness. Among the architectures, EfficientNetB0—trained with Geometric & Color augmentation and selected via stratified 3‑fold cross‑validation—achieved optimal classification performance on cropped images.
Despite the constraints posed by class imbalance and a limited dataset size, the work confirms the viability of deep learning for automating rabies diagnosis. The authors deployed an online tool to facilitate practical access, establishing a framework that can be extended to other medical imaging applications.
This research underscores the potential of optimized deep‑learning models to transform rabies diagnostics and improve public health outcomes.
Reference: Rabies diagnosis in low‑data settings: A comparative study on the impact of data augmentation and transfer learning (arXiv:2604.19823v1). https://arxiv.org/abs/2604.19823
Source: Rabies diagnosis in low-data settings: A comparative study on the impact of data augmentation and transfer learning
Domain: arxiv.org
Comments load interactively on the live page.