Pathological micrographs play a crucial role in diagnostics, particularly in cancer screening and pathological analysis. However, factors such as poor lighting or focus errors during image acquisition often lead to blurring, which compromises diagnostic accuracy and efficiency. This paper proposes an efficient model for microscopic pathological images deblurring based on diffusion, named PDDiff. In PDDiff, the image deblurring process is guided by a low-dimensional latent feature representation, which is generated through multi-scale approach and an improved diffusion model. Additionally, we design a Targeted Multi-Head Sparse Attention Module and a Neighborhood Feed-Forward Network to efficiently remove blur and enhance detail restoration. Meanwhile, we propose a Lightweight Denoising Diffusion Implicit Model, which significantly improves computational efficiency by adopting a lightweight encoder structure and an efficient diffusion process. Extensive experiments on multiple datasets demonstrate that our method outperforms existing state-of-the-art approaches, achieving 37.32 dB PSNR and 0.9703 SSIM on the Landing dataset, and 39.44 dB PSNR and 0.9891 SSIM on the CRC-VAL-HE-7K dataset, offering a robust and efficient solution to deblurring pathological images and extending the use of diffusion models in medical imaging.