Recent advancements demonstrate that implicit neural representations of the Signed Distance Field (SDF) yield remarkable results in reconstructing shapes from freehand 3D ultrasound imaging. Typically, multi-view data is arranged in rows and columns, but fusing ultrasound data or networks remains a significant challenge. To address this issue, we propose a novel neural signed distance function for ultrasound shape reconstruction, guided by a newly introduced gated fusion and Sampson distance, termed GS-SDF. The core of our method lies in combining row-SDF and column-SDF networks using an adaptive gating mechanism and optimizing a fusion SDF network with a Sampson loss. Specifically, for row- and column-scanned data, we first employ implicit neural representations to model their SDFs. We then design a gating mechanism to dynamically assign weights to the row-SDF and column-SDF networks, significantly enhancing the fusion process and enabling more accurate fitting during training. Additionally, we introduce the Sampson distance to improve the accuracy of distance computation between points and the surface, compared to conventional algebraic methods. This approach provides a more faithful evaluation of the loss between scanned data and model predictions. We demonstrate the efficacy of GS-SDF on four benchmark datasets acquired using ultrasound transducer probes and computed tomography, achieving state-of-the-art performance compared to the competing reconstruction approach. The code for our method will be made publicly available.