MBANet

Segmenting kidneys in ultrasound images is difficult due to variable anatomy, heterogeneous structure, and inconsistent image quality. The study introduces MBANet, a deep learning architecture that combines multi-scale feature extraction, multi-branch encoders, and a master decoder to preserve detailed image information and improve segmentation accuracy. Using a step-by-step training strategy, MBANet outperformed existing methods across six metrics, achieving high pixel accuracy, IoU, precision, recall, specificity, and F1 score on kidney ultrasound datasets.