The paper introduces a novel attention gate (AG) model for medical image analysis, which automatically learns to focus on target structures of varying shapes and sizes. AGs are integrated into standard CNN models, such as VGG or U-Net, to improve model sensitivity and prediction accuracy while reducing computational overhead. The proposed AG models are evaluated on various tasks, including medical image classification and segmentation. For classification, the AGs are applied to real-time fetal ultrasound scan plane detection, demonstrating improved classification performance and efficient object localization. For segmentation, the Attention U-Net architecture is extended to handle 3D CT abdominal datasets, showing consistent improvements in prediction accuracy across different datasets and training sizes. The attention mechanism also provides fine-grained attention maps that enhance interpretability. The source code for the proposed AG models is publicly available.The paper introduces a novel attention gate (AG) model for medical image analysis, which automatically learns to focus on target structures of varying shapes and sizes. AGs are integrated into standard CNN models, such as VGG or U-Net, to improve model sensitivity and prediction accuracy while reducing computational overhead. The proposed AG models are evaluated on various tasks, including medical image classification and segmentation. For classification, the AGs are applied to real-time fetal ultrasound scan plane detection, demonstrating improved classification performance and efficient object localization. For segmentation, the Attention U-Net architecture is extended to handle 3D CT abdominal datasets, showing consistent improvements in prediction accuracy across different datasets and training sizes. The attention mechanism also provides fine-grained attention maps that enhance interpretability. The source code for the proposed AG models is publicly available.