Accurate attenuation correction (AC) is essential for quantitative brain positron emission tomography (PET), but it presents a significant challenge for dedicated PET-only systems that lack anatomical information from computed tomography (CT) or magnetic resonance imaging. This review provides a comprehensive overview of AC methods designed for stand-alone brain PET. We categorise and analyse existing approaches, including traditional segmentation-based techniques, atlas- and template-driven strategies, methods for simultaneous reconstruction of attenuation and activity, and modern data-driven deep learning techniques. We discuss the inherent trade-offs of each category: template-based methods are computationally efficient and robust for normal anatomy but may fail in atypical cases, while simultaneous reconstruction methods can be powerful but are often sensitive to noise and require time-of-flight information. Deep learning approaches have demonstrated the highest potential for generating patient-specific attenuation maps but are critically dependent on large, diverse training datasets and face challenges in generalisability. Finally, we survey the methods currently implemented in commercial brain-dedicated PET systems, revealing a gap between cutting-edge research and clinical practice, where simpler, more established techniques are still preferred. This review highlights the ongoing need for CT-less AC solutions that balance accuracy, robustness, and practical feasibility for clinical adoption.
Jehl et al. (Sun,) studied this question.