Application and integration of deep learning in tumour radiomics: bibliometrics and visualisation analysis.

INTRODUCTION: The integration of deep learning with tumour radiomics represents a significant advance in precision oncology, offering a powerful alternative to traditional radiomics that relies on manually designed imaging features and is limited in capturing tumour complexity. METHODS: To analyze integration strategies, trends, and future directions in this field, a bibliometric analysis was conducted using the Web of Science Core Collection, retrieving 4,915 relevant articles as of October 28, 2025. Data were examined with CiteSpace, VOSviewer, and bibliometrix, employing co-authorship, co-citation, keyword analysis, Latent Dirichlet Allocation topic modeling, and burst detection. RESULTS: Annual publications rapidly increased from 253 in 2019 to 1,021 in 2024, with China and the United States as leading contributors and the U.S. serving as the central hub for international collaboration. Eighteen core topics were identified and grouped into four domains: tumour characterization and diagnosis, treatment response and prognosis, molecular profiling, and methodological innovation. The field has evolved from early texture analysis toward current focuses on multimodal fusion, molecular subtyping, and deep learning-driven predictive modeling. DISCUSSION: This bibliometric analysis suggests a clear transition in oncological image analysis from descriptive morphology toward predictive and prognostic modeling, driven by the deep integration of deep learning with radiomics. Future efforts should prioritize multinational collaboration, methodological standardization, and prospective clinical validation to help bridge the gap toward clinical integration, while acknowledging that reproducibility and other validation challenges persist.