Enhanced Segmentation of High-Grade and Low-Grade Brain Tumors Using Advanced 3D U-Net++ with Hybrid Lion-Spider Monkey Optimization

Abstract

Appropriate segmentation of brain tumors from MRI images is crucial for accurate evaluation and treatment management. This paper presents an enhanced approach to segment high- and low-grade gliomas by optimizing the 3D U-Net++ architecture using the Hybrid Lion-Spider Monkey Optimization Algorithm (LSMA). The LSMA integrates the Lion-Spider Monkey Algorithm (LSMA) to improve parameter tuning and feature extraction, significantly enhancing the segmentation process. The study utilizes the BRATS 2020 dataset, which includes T1-weighted, T2-weighted, and FLAIR MRI scans, capturing the distinctive features of the tumors. Preprocessing steps involve estimating image noise tiers and applying the Frost clear out to reduce clutter even as keeping essential details. The modalities are blended into a unified dataset and standardized to make sure regular depth throughout images. Data augmentation strategies, such as rotation and deformation, are employed to increase set of rules resilience. In terms of network structure, the 3-D U-Net++ model features an encoder-decoder shape with dense connections for effective information transmission and characteristic extraction. Deep supervision with auxiliary outputs similarly refines gradient float and improves segmentation accuracy. The model is to start with pretrained on downscaled images to capture large-scale capabilities, accompanied via first-class-tuning on complete-decision pictures for stronger aspect detail. Evaluation on a separate test set demonstrates that the LSMA-optimized 3-D U-Net++ achieves an outstanding accuracy of 99%, surpassing previous methods. This advanced architecture, applied in Python, gives a fairly correct and flexible answer for brain tumor segmentation, offering precious support for clinical practitioners in making informed remedy choices and planning.