DUAL PROMPTING FOR DIVERSE COUNT-LEVEL PET DENOISING.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Image Processing

Background:

• Positron emission tomography (PET) imaging is challenged by diverse count levels in acquired volumes.
• Existing denoising models struggle to generalize across these varying noise conditions.
• Unified models are needed to address the heterogeneity of PET data.

Purpose of the Study:

• To develop a generalizable PET denoising method using prompt learning for diverse count levels.
• To create a unified denoising model adaptable to different PET data quality.
• To enhance the performance of PET denoising through novel prompt-based strategies.

Main Methods:

• Proposed a dual prompt learning strategy: an explicit count-level prompt and an implicit general denoising prompt.
• Developed a prompt fusion module to integrate heterogeneous prompts.
• Implemented a prompt-feature interaction module to guide the denoising process dynamically.
• Evaluated the method on 1940 low-count 3D PET volumes from ¹⁸F-MK6240 tau PET studies.

Main Results:

• The dual prompting approach demonstrated improved PET denoising performance.
• The unified model, guided by prompts, effectively handled varying PET count levels.
• Outperformed traditional count-conditional denoising models in low-count scenarios.
• Personalized prompts allowed for deployment across different PET cases.

Conclusions:

• Dual prompt learning offers a powerful and generalizable solution for PET denoising across diverse count levels.
• The proposed method enables efficient training of a unified model adaptable to various PET imaging conditions.
• This approach advances the application of AI in medical imaging for improved image quality and diagnostic accuracy.