Tumors and lesions in medical images often exhibit intricate textures and ambiguous boundaries, posing challenges for conventional object detection frameworks. To overcome these issues, we introduce an enhanced YOLOv8 architecture tailored for medical object detection, designed to improve the network’s capacity for capturing fine-grained structural details. Our architecture introduces two novel components to enhance both local texture representation and global context modeling. First, the Fourier Re-parameterized Convolution applies a forward Fourier transform to project spatial features into the frequency domain, performs spatial convolutions in this transformed space, and then applies an inverse Fourier transform to return to the spatial domain. This process enables explicit integration of complementary spatial- and frequency-domain information, improving the network’s ability to capture periodic patterns and fine structural details. Second, the Re-parameterized C2f module employs branch re-parameterization to merge multiple convolutional pathways into a single equivalent structure at inference, thereby enriching cross-stage feature interactions while maintaining computational efficiency. We evaluate the proposed model on four medical object detection benchmarks: BR35H, ISIC2018, ACDC, and a liver tumor dataset derived from LiTS. Across all datasets, it consistently surpasses the baseline YOLOv8 on multiple evaluation metrics, with especially pronounced improvements for low-contrast and poorly delineated targets, while preserving competitive inference speed.