This article presents a curated dataset of 4,200 B-mode ultrasound images of human forearms, intended for use in deep learning-based image reconstruction and segmentation tasks. Data have been collected using the Verasonics Vantage 64LE system equipped with an L11-5v linear array transducer, known for its high spatial resolution. Each scanning session captures 100 frames per orientation across both arms of 14 healthy subjects, covering multiple anatomical views. The dataset provides grayscale B-mode images stored in MATLAB .mat format and is publicly available through Mendeley Data. This dataset is valuable for researchers in medical image analysis, especially those developing deep learning models for enhancing ultrasound imaging quality and anatomical structure segmentation.