MGTP: Multi-Granularity Textual Prompts for Low-Dose Brain PET Image Denoising via Adversarial Diffusion Model

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Nuclear Medicine

Background:

• Positron Emission Tomography (PET) is crucial in clinical settings but faces challenges with radiation risks at standard doses and poor quality at low doses.
• Current low-dose PET denoising methods often neglect non-image textual data, leading to suboptimal results with lost context and detail.

Purpose of the Study:

• To develop an advanced method for denoising low-dose PET images by integrating textual information.
• To improve the quality and clinical utility of low-dose PET scans, thereby reducing radiation exposure.

Main Methods:

• Proposed Multi-Granularity Textual Prompts (MGTP) using an adversarial diffusion model to denoise low-dose PET images.
• Introduced a Cross-Modality Selective Conditioning (CMSC) module to harmonize image and multi-granularity textual prompts.
• Developed a Masked Prompt Reconstruction Network (MPR-Net) to preserve semantic and detailed information in denoised images.

Main Results:

• The MGTP method effectively denoises low-dose PET images by incorporating diverse textual information.
• The CMSC module successfully integrates semantic contexts and degradation details from textual prompts.
• MPR-Net mitigated distortions, preserving crucial image semantics and details.

Conclusions:

• The proposed MGTP method achieves state-of-the-art performance in low-dose PET image denoising.
• Integrating multi-granularity textual prompts significantly enhances denoising quality compared to image-only methods.
• This approach offers a promising solution for high-quality, low-dose PET imaging in clinical practice.