📄 中文摘要
正电子发射断层扫描（PET）图像在医学诊断中至关重要，提供人体内分子过程的详细信息。然而，PET图像常受复杂噪声影响，这会掩盖关键诊断信息。PET图像质量受多种因素影响，包括扫描仪硬件、图像重建、示踪剂特性、剂量/计数水平以及采集时间。在PET图像去噪方面，扩散模型已显示出其在生成高保真图像方面的强大能力。然而，现有扩散模型在PET去噪中面临主要挑战：它们通常采用无条件生成或单一条件（如CT图像或低计数PET图像）输入。这种单一条件限制了模型对不同噪声类型和程度的适应性。为解决此问题，一种新颖的文本控制PET去噪框架被提出。
📄 English Summary
Text controllable PET denoising
Positron Emission Tomography (PET) imaging is a vital tool in medical diagnostics, offering detailed insights into molecular processes within the human body. However, PET images often suffer from complicated noise, which can obscure critical diagnostic information. The quality of the PET image is impacted by various factors including scanner hardware, image reconstruction, tracer properties, dose/count level, and acquisition time. Diffusion models have demonstrated strong capabilities in generating high-fidelity images, making them promising for PET image denoising. Nevertheless, existing diffusion models in PET denoising typically employ unconditional generation or rely on a single conditional input, such as CT images or low-count PET images. This limited conditioning restricts the models' adaptability to diverse noise types and magnitudes. To address this, a novel text-controllable PET denoising framework is proposed. The framework leverages the flexibility and expressiveness of text prompts, allowing users to describe desired denoising effects or regions of interest in natural language. For instance, a user might input “remove background noise while preserving tumor edge details” or “denoise low-dose PET images.” By encoding text prompts as conditional inputs, the diffusion model learns more refined and controllable denoising strategies. The core of this method involves fusing text embeddings with PET image features to guide the diffusion process toward generating denoised images that align with the text descriptions. This enables the denoising process to not only remove noise but also to be customized according to clinical needs, thereby enhancing diagnostic accuracy.