An Efficient Brain Tumor Segmentation and Classification Framework Using Recurrent Unet + + and Adaptive Inception Net with Exponential Tanh Function

Abstract

Globally, brain cancer disease classification is one of the complex challenges for medical treatment and diagnosis using clinical image analysis. The existing techniques faced problems in handling unclear boundaries and critical tumor shapes, resulting in outcome variability. Often, these models depend on handcrafted features, which do not have the tendency to detect patterns and subtle differences involved in brain cancer images. Hence, it leads to poor capability to efficiently classify some brain tumor types. To overcome such drawbacks, a novel automated deep learning approach is suggested for brain cancer classification from MRI images. Here, the available public MRI database is utilized to garner raw images. The gathered raw images are subjected to Recurrent Unet[Formula: see text] with Spatial-Temporal Attention (RUnet[Formula: see text]-STA) for the segmentation process. This particular network is integrated using the strengths of Unet[Formula: see text] with recurrent layers to capture spatial-temporal features effectively, leading to enhanced tumor segmentation performance. Further, for brain tumor classification, the segmented images are fed into the Adaptive InceptionNet with ExponentialTanhFunction (AINet-ET). By fine-tuning InceptionNet parameters using the Reformulated Random Parameter Sculptor Optimization Algorithm (RRPSO), the suggested approach achieves superior classification accuracy and demonstrates increased robustness in identifying brain tumors. A well-optimized model is more robust in the classification of brain tumors. Finally, the efficiency of the proposed method is analyzed through a comparison with traditional methods.