Thoracic malignancies, driven primarily by lung cancer, are the leading cause of cancer-related death worldwide. Artificial intelligence (AI) has shown considerable potential in thoracic oncology, particularly in medical imaging-based applications that improve diagnosis and prognosis. Although many studies have explored diverse AI models across various imaging modalities, most studies remain retrospective, with few prospective or qualitative assessments involving end-users. To assess the current landscape of translational imaging-based AI research in thoracic malignancies, we conducted a scoping review of post-diagnostic models to predict treatment response, patient survival, and precise tumor staging and identified 19 eligible studies that demonstrated potential clinical applicability with external validation. This review summarizes recent advances, as well as major challenges related to data management, model evaluation, and generalizability across clinical settings. Based on these findings, we propose recommendations for robust validation frameworks, standardized reporting of methods for replication, and responsible integration of AI tools into the clinic. Adoption of these recommendations may help bridge the gap between promising models and clinical integration, leading to a meaningful impact on patient care.
Milantoni et al. (Thu,) studied this question.