Over the last decade, the introduction of immune checkpoint inhibitors (ICIs) has profoundly transformed the therapeutic landscape. As a result, novel response patterns have emerged, typically reflecting the unique effects of these therapies, which recruit immune cells into the tumor microenvironment.1 These new patterns include pseudoprogression, hyperprogression and dissociated response (DR).2 Pseudoprogression is defined as the appearance of additional lesions or an increase in tumor size, which would traditionally be classified as disease progression, but subsequently showing shrinkage or stabilization on later imaging examinations, due to the temporary infiltration of cytotoxic T lymphocytes into tumors.3 On the other hand, in patients whose tumor volume increased at an exceptionally fast rate during ICI therapy, far surpassing the growth observed before treatment, this phenomenon was classified as hyperprogressive disease.4 Dissociated responses are characterized by simultaneous occurrence of tumor progression and regression across different sites; it is rare and recently recognized, with reported rates varying from 3.3% to 47.8% depending on definitions and cancer types2 (Fig. 1).

Within this framework, it becomes crucial to identify which patients are most likely to derive sustained benefit from immune checkpoint blockade and which, conversely, may experience pseudoprogression or even detrimental hyperactivation of the immune system, ultimately leading to unfavorable outcomes. Therefore, deeper understanding of these divergent response patterns, and biological characteristics of the disease are essential to optimize patient selection and guide individualized therapeutic strategies.

Fluorine-18-fluorodeoxyglucose ([18F]FDG) positron emission tomography/computed tomography (PET/CT) is widely recognized as a fundamental imaging modality for both disease staging and treatment response evaluation across a broad range of malignancies. Typically, response criteria according to PET Response Criteria in Solid Tumors (PERCIST 1.0) include complete response (CR), partial response (PR), stable disease (SD), or progressive disease (PD).5,6

In recent years, radiomics has emerged as a rapidly evolving field centered on the quantitative assessment of medical imaging data. By further exploiting machine learning techniques, it enables the development of predictive models capable of capturing complex imaging patterns.7 The fundamental objective of radiomics is to extract quantitative parameters, known as features, from medical images, and associate them with relevant clinical outcomes such as therapeutic response, disease progression, or patient survival.

Within this framework, both preclinical and clinical investigations have highlighted the promising role of PET-based radiomics in predicting the efficacy of immunotherapy.8 Regarding the prognostic stratification of patients prior to immunotherapy, [18F]FDG PET/CT has shown promising preliminary findings, as several PET-derived metrics, such as whole-body metabolic tumor volume (wbMTV) and total lesion glycolysis (wbTLG), have been identified as potential predictors of treatment response.9, 10, 11

The aim of this review is to provide a comprehensive synthesis of the current literature on the use of PET radiomics for predicting response to immunotherapy in cancer patients, as well as to discuss the main challenges emerging from data analysis and propose potential directions for its broader integration into clinical practice.