Currently, advanced two-stage object detectors heavily rely on time-consuming oriented proposal generation methods, which significantly limit detection speed and constitute a major bottleneck in oriented detector development. To address this, we propose E2E-LAD, an end-to-end detection algorithm based on diffusion models, achieving high performance through one-stage training. E2E-LAD integrates parameter-efficient fine-tuning with low-rank adaptation, updating model weights efficiently by adding or multiplying low-rank matrices to the original weights, enabling rapid adaptation to new data distributions. Furthermore, oriented object detection is reformulated as a denoising diffusion process from noisy boxes to target boxes, allowing dynamic box quantity adjustment and iterative refinement, thereby enhancing flexibility and accuracy. To address the angular periodicity problem overlooked by existing encoding methods, a unit-periodic constraint is introduced into the midpoint offset representation, along with a unit-circle-based loss, significantly improving angle prediction accuracy. Experiments show that E2E-LAD achieves mAP of 77.87%, 72.52%, and 66.95% on DOTA-v1.0, DOTA-v1.5, and DIOR-R respectively. These results validate the effectiveness and superiority of E2E-LAD in oriented object detection and provide new insights for further research.