Abstract Accurate brain tumor segmentation and classification are critical for effective clinical diagnosis and treatment planning. Nevertheless, the current deep learning solutions commonly face the problem of tumor heterogeneity, poor contextual modeling, and ineffective discriminating of features. In response to these issues, this paper will present a new multi‐modal brain tumor detection model, which will combine noise‐conscious pre‐processing Difference of Gaussian Square Kernel Filtering (D‐GSKF), fine‐tuning tumor segmentation, U‐Net, Efficient Long‐Range Attention Network (U‐Net‐ELAN), feature extraction using Point Transformer (PiT) and Siamese Neural Networks, which are optimized through the Walrus Optimization Algorithm (SNN‐WaOA). This work is important because it introduces a novel synergistic combination of long‐range attention‐inspired segmentation, transformer‐inspired spatial feature modeling, and bio‐inspired optimization to similarity‐based tumor classification. The effectiveness of the suggested framework is supported by the large number of experiments performed on four benchmark data sets, including Figshare, BraTS 2020, Harvard, and Brain MRI, with the highest accuracy rate of 99.9%, the highest recall of 99.8%, and the lowest error rate of 0.1%, surpassing the results of a number of state‐of‐the‐art methods. These findings suggest that the suggested method can be a stable and clinically useful method of automated brain tumor detection in various MRI modalities.
Padma et al. (Wed,) studied this question.