Interpretable Sperm Morphology Classification via Attention-Guided Deep Learning
Researchers have developed an attention-guided deep learning framework to improve the interpretability and accuracy of sperm morphology classification. By integrating a pre-trained EfficientNet-B0 model with a Convolutional Block Attention Module (CBAM), the system effectively focuses on critical sperm head features. This approach achieved high accuracy rates of 90.2% and 93.9% on public datasets, surpassing simpler models and providing visual explanations for its classifications. AI
IMPACT This research offers a more transparent and accurate AI tool for clinical applications in fertility analysis.