Segmentation of Brain Tumor from Magnetic Resonance Imaging Using Handcrafted Features with BOA-Based Transformer

Abstract

Early brain tumor detection is crucial for improving patients’ prognosis and chances of survival. Physical analysis of brain tumor magnetic resonance imaging (MRI) images is necessary for this task. Consequently, computational techniques are required for more precise tumor diagnosis. However, evaluations of shape, volume, boundaries, size, tumor identification, segmentation, and classification remain challenging. Additionally, characteristics of cancer, such as fuzziness, complex backgrounds, and significant variations in size, shape, and intensity distribution, make accurate segmentation challenging. This work suggests a novel Optimizer-based Semantic-Aware Transformer (OSAT) for brain tumor segmentation in order to address these problems. Moreover, MRI data was manually analyzed to extract features based on texture, intensity, and other factors. The Bonobo optimization algorithm (BOA) improves SAT and increases feature representation learning capabilities with less memory and computational complexity. Several evaluation metrics were used in this work to assess performance on the three Brain Tumor Segmentation (BraTS) challenge datasets, including segmentation measures. By enhancing OSAT’s performance with the addition of handcrafted features, a more reliable and broadly applicable solution was also achieved. This study may have significant applications in the field of accurate and efficient brain tumor segmentation. Future studies could examine various feature fusion techniques and incorporate additional imaging modalities to improve the efficacy of the proposed method.