Abstract
INTRODUCTION: Brain tumors pose significant harm to the functionality of the human nervous system. There are lots of models which can classify brain tumor type. However, the available methods did not pay special attention to long-range information, which limits model accuracy improvement. METHODS: To solve this problem, in this paper, an enhanced short-range and long-range dependent system for brain tumor classification, named as EnSLDe, is proposed. The EnSLDe model consists of three main modules: the Feature Extraction Module (FExM), the Feature Enhancement Module (FEnM), and the Classification Module. Firstly, the FExM is used to extract features and the multi-scale parallel subnetwork is constructed to fuse shallow and deep features. Then, the extracted features are enhanced by the FEnM. The FEnM can capture the important dependencies across a larger sequence range and retain critical information at a local scale. Finally, the fused and enhanced features are input to the classification module for brain tumor classification. The combination of these modules enables the efficient extraction of both local and global contextual information. RESULTS: In order to validate the model, two public data sets including glioma, meningioma, and pituitary tumor were validated, and good experimental results were obtained, demonstrating the potential of the model EnSLDe in brain tumor classification.