The occurrence of cancer metastasis typically indicates the spreading of cancer cells to other tissues or organs, leading to a worsening of the disease. Identifying whether cancer has metastasized can help doctors infer the progression of a patient's condition and is an essential prerequisite for devising treatment plans. Fluorine 18 fluorodeoxyglucose positron emission tomography / computed tomography (18F-FDG PET/CT) is an advanced cancer diagnostic imaging technique that provides both metabolic and structural information. In the task of identifying cancer metastasis, effectively integrating metabolic and structural information is one of the key technologies to improve feature representation and recognition performance. Addressing these issues, this paper proposes a cancer metastasis identification network model using dynamic coordinated metabolic attention and structural attention. Specifically, metabolic and structural features are extracted by adding a dynamic coordinated attention module (DCAM) to two branches of ResNet networks, combining the high metabolic spatial information from PET images with the texture structure information from CT images, and dynamically adjusting this process through iterations. Next, to improve the efficacy of feature expression, a multi-receptive field feature fusion module (MRFM) is included in order to execute multi-receptive field fusion of semantic features. To validate the effectiveness of our proposed model, experiments were conducted on a private lung lymph nodes dataset and a public soft tissue sarcomas dataset. The accuracy of our method was 76.0% and 75.1% for the two datasets, respectively, showing an improvement of 6.8% and 5.6% compared to ResNet, indicating the effectiveness of our method.

BACKGROUND: Bibliometrics has been applied to the study of tumor image segmentation, which can indicate the current research hotspots and trends. METHOD: In this study, bibliometric analyses were performed on data retrieved from the Web of Science database. A total of 3024 articles on the application of tumor image segmentation from January 1, 2003 to March 31, 2024 were analyzed for the characteristics of the articles, including the number of year publications, country/region, institution, journal, author, keywords, and references. Visualising co-authorship, co-citation and co-occurrence analysis with VOSviewer RESULTS: The annual publication volume of tumor image segmentation literature shows that from the first time of more than 100 articles in 2016, the publication volume of literature in this field has achieved a surge, reaching 571 articles by 2023. Mainland China ranked first in terms of publication volume (n=1149) and average publication year (n=2020.93). IEEE Transactions on Medical Imaging was the journal with the highest publication volume and average number of citations. Chinese Academy of Sciences (n=67) was the most prolific institution, and Harvard University was the most prestigious with a total number of citations and an average number of citations of 3015 and 215, respectively. In terms of keywords, co-occurrence analysis of 96 keywords with a frequency of more than 30 times produced four clusters (1) methods of image segmentation (2) applications of image segmentation (3) applications of image segmentation (4) image segmentation modelled on MRI; transformer, attention mechanism, u-net are the latest keywords. The analysis of keywords helps scholars to understand and identify the current research hotspots and research directions. CONCLUSION: Within the last 20 years, the number of articles on the application of tumor image segmentation has increased year by year. It presents a strong connection between countries/regions and authors. This study shows the global trends, research hotspots and emerging topics in this field and reviews some of the knowledge literature about tumor image segmentation applications from past studies. This will provide good research guidelines for researchers in this field.