Brain Tumor Segmentation using Osprey Optimization Assisted DeepLabV3+ Model

Brain tumors are among the frequently diagnosed malignant conditions across all age groups. Accurately determining their grade has a major challenge for radiologists in clinical assessments and automatic diagnostic systems. Identifying tumor types and implementing preventive measures remains one of the most complex processes of brain tumor classification. Various Deep Learning (DL) models are proposed in the existing approaches for enhancing the accuracy of brain tumor classification. But, the challenges like training time and complexity are occurred in these works. To tackle these issues, this work presents a Enhanced DeepLabV3+ to segment and categorize brain tumor. At first, non-local means (NLM) filtering is utilized for pre-processing for reducing noise and preserving essential structural details. Then, the Enhanced DeepLabV3+ is employed for segmentation, with AlexNet is the backbone for segmentation tasks. To further refine the segmentation process, hyperparameter optimization of the DeepLabV3 architecture is conducted using the Osprey Optimization Algorithm (OOA) approach and provide significant improvements in brain tumor segmentation performance. The evaluation is performed on the Brain TCIA and Figshare datasets and achieved better accuracies of 98.97% and 99.23% respectively.