Abstract
Perineural invasion (PNI) is a marker of aggressive behavior and poor prognosis for gastrointestinal (GI) malignancies. Research on PNI in tumors has evolved from initially describing its pathological features to recently exploring the interaction between tumors and nerves. This study is entirely based on data obtained from the Web of Science Core Collection and conducted a bibliometric analysis of 196 articles published in the database from 1993 to 2025. Using VOSviewer and CiteSpace software, we quantified the outputs, drew a collaboration network diagram, and analyzed co-citations, keyword co-occurrences/clustering, and outbreak situations. The number of annual publications after 2013 has significantly increased, with the greatest contributions coming from China and the United States. A comprehensive analysis of the existing literature has yielded three main themes: first, the radiomics prediction of PNI; second, the prognostic value of PNI for recurrence/survival; third, the mechanism of interaction between nerves and tumors. The emergence of terms such as "radiomics", "machine learning", and "deep learning" marks a paradigm shift towards an artificial intelligence (AI)-driven, non-invasive risk stratification method. These transitions highlight the translational potential of integrating AI-based models and insights into the mechanisms of tumor-neuron interaction and applying them to clinical practice. This study depicts the evolution of PNI research from pathological identification to tumor-neuron interaction, reveals the integration of oncology and neuroscience in GI malignancies, and highlights emerging directions. Future work should focus on the prospective construction of AI tools and multi-center multimodal validation, deeper mechanism analysis, and the inclusion of underrepresented cancer subtypes to provide a roadmap for the development of precision oncology for GI malignancies.