CT-Less TOF PET: challenges and innovations for quantitative imaging.

Area of Science:

• Medical Imaging
• Nuclear Medicine
• Artificial Intelligence

Background:

• Time-of-Flight (TOF) PET imaging inherently corrects for physical degradation, including attenuation, a task traditionally requiring CT scans.
• Early advancements in PET reconstruction, like Maximum-Likelihood reconstruction of Attenuation and Activity (MLAA), focused on inherent self-correction.
• Current research prioritizes Deep Learning (DL) for generating attenuation maps from PET data, moving beyond traditional methods.

Purpose of the Study:

• To evaluate the shift in focus from TOF PET's self-correction capabilities to Deep Learning-based attenuation map generation.
• To explore the potential of DL models (CNNs, GANs, diffusion models) in synthesizing accurate attenuation maps from suboptimal PET images.
• To discuss the implications of DL in medical imaging and its comparison to traditional, human-developed AI models.

Main Methods:

• Review of advancements in PET imaging reconstruction and attenuation correction techniques over the past decade.
• Exploration of Deep Learning approaches, including Convolutional Neural Networks (CNNs), Generative Adversarial Networks (GANs), and diffusion models, for image-to-image tasks.
• Analysis of the challenges and potential of generating attenuation maps using DL from non-attenuation-corrected PET images, considering data availability and potential misalignments in PET-CT scans.

Main Results:

• A significant shift in research focus from TOF PET's inherent self-correction to Deep Learning-based attenuation map synthesis.
• Demonstration of DL models' capability to perform image-to-image translation for generating attenuation maps.
• Identification of the key question regarding the level of detail recoverable from suboptimal PET images using DL.

Conclusions:

• Deep Learning has become a dominant approach for generating attenuation maps in PET imaging, transforming the field.
• The power of DL in image synthesis raises questions about the future role of traditional AI and human expertise in medical imaging.
• Further research and time are needed to fully understand the long-term impact and capabilities of DL in PET attenuation correction.