Artificial intelligence in oncological positron emission tomography: advancing image analysis and interpretation

Functional and metabolic information provided by positron emission tomography (PET) imaging, such as patient diagnosis, tumour staging, and treatment evaluation, plays an important role in the clinical management of patients with cancer. Nonetheless, its clinical efficacy may be inhibited by differences in image quality and limitations in quantitative robustness. Artificial intelligence (AI) has transformed oncological PET imaging by improving image quality and facilitating a more consistent extraction of quantitative metrics. Recent research emphasises the value of AI in improving diagnostic accuracy and prognostic modelling. However, to ensure that AI-based PET analysis is successfully implemented in clinical practice, challenges such as imaging data standardisation, the development of reliable explainability methods, and the establishment of regulatory frameworks must be addressed. To optimise individualised care, future progress will likely be based on multimodal integration, federated learning, and probabilistic deep learning. Overall, this review highlights both the current progress and the remaining challenges of AI in oncological PET, aiming to provide a balanced perspective for future clinical translation.