We propose OCTDiff, a bridged diffusion model with Adaptive Noise Aggregation and Multi-Scale Cross-Attention, trained with quality-guided loss. It achieves SOTA enhancement on portable OCT images and shows potential for broader clinical use.