ARTIFICIAL INTELLIGENCE IN NUCLEAR ONCOLOGY: A REVIEW OF CURRENT EVIDENCE AND TRANSLATIONAL PRIORITIES

Abstract

Artificial intelligence (AI) is fundamentally reshaping the practice of nuclear medicine in oncological diagnosis. This review systematically evaluates the technological foundations, clinical applications, diagnostic performance, and barriers to clinical translation of AI in nuclear medicine for cancer diagnosis. A comprehensive literature search was conducted across PubMed, Scopus, and Web of Science databases for studies published between 2020 and 2025. Studies were categorized by AI methodology (radiomics, deep learning, large language models) and clinical application (diagnostic classification, treatment response prediction, prognosis, image enhancement, dosimetry). Current evidence demonstrates that AI-driven radiomics models achieve significant performance improvements in diagnostic classification across multiple tumor types, while deep learning assistance enhances clinician diagnostic sensitivity and accuracy. In radiopharmaceutical therapy, AI-supported individualized dosimetry strategies show potential for simplified imaging protocols. Large language model applications in nuclear medicine expanded from 41 tasks in 2023 to 434 tasks in 2025. However, insufficient data standardization, lack of external validation, limited model interpretability, and unclear regulatory pathways remain significant barriers to clinical translation. Future research should focus on prospective multicenter validation, standardized imaging protocol development, and explainable AI frameworks.