Bulletin of Electrical Engineering and Informatics Vol. 14, No. 2, April 2025, pp. 1447~1455 ISSN: 2302-9285, DOI: 10.11591/eei.v14i2.8730  1447 Journal homepage: http://beei.org Glioma segmentation using hybrid filter and modified African vulture optimization Bhagyalaxmi Kuntiyellannagari 1 , Bhoopalan Dwarakanath 2 1 Department of Computer Science and Engineering, Faculty of Engineering and Technology, SRM Institute of Science and Technology, Ramapuram, Chennai, India 2 Department of Information Technology, Faculty of Engineering and Technology, SRM Institute of Science and Technology, Ramapuram, Chennai, India Article Info ABSTRACT Article history: Received May 23, 2024 Revised Oct 9, 2024 Accepted Nov 19, 2024 Accurate brain tumor segmentation is essential for managing gliomas, which arise from brain and spinal cord support cells. Traditional image processing and machine learning methods have improved tumor segmentation but are often limited by accuracy and noise handling. Recent advances in deep learning, particularly using U-Net and its variants, have achieved significant progress but still face challenges with heterogeneous data and real-time processing. This study introduces a hybrid bilateral mean filter for noise reduction coupled with an ensemble deep learning model that integrates U- Net, InceptionV2, InceptionResNetV2, and W-Net to enhance segmentation accuracy and efficiency. Additionally, we propose a novel modified African vulture optimization algorithm (MAVOA) to further refine segmentation performance. Evaluated on the BraTS 2020 dataset, our model achieved a loss of 0.023 with strong performance metrics: 98.2% accuracy, 97.2% mean intersection over union (IOU), and 99.1% precision. It effectively segmented glioma subregions with dice scores of 0.96 for necrotic areas, 0.97 for edema, and 0.91 for enhancing regions. On the BraTS 2021 dataset, the model maintained high accuracy 96.4%, mean IOU 95.9%, and dice coefficients of 0.91 for necrotic areas, 0.95 for edema, and 0.92 for enhancing regions. Keywords: Deep learning Glioma Inception ResNetV2 Medical imaging Segmentation U-Net W-Net This is an open access article under the CC BY-SA license. Corresponding Author: Bhagyalaxmi Kuntiyellannagari Department of Computer Science and Engineering, Faculty of Engineering and Technology SRM Institute of Science and Technology Ramapuram, Chennai, India Email: bk8019@srmist.edu.in 1. INTRODUCTION Glioblastoma is the most common and aggressive type of primary brain tumor. Statistics reveal that 85-90% of all primary central nervous system (CNS) tumors are brain tumors. Newly diagnosed cases of brain and CNS cancer are responsible for about 3% of all cancers globally. In European countries, these cases are five times higher than in Asian countries. Early-stage diagnosis and treatment require automatic segmentation of brain tumors, which is expensive and time-consuming if done manually. Recent advancements in image processing and computer vision have significantly contributed to this area. Gliomas, a type of brain tumor, are categorized into low-grade glioma (LGG), which grows slowly, and high-grade glioma (HGG), which can be life-threatening. According to the World Health Organization, HGG tumors are critical, with a maximum survival rate of two years, while LGG-affected patients can have several years of life expectancy [1]. Despite advancements in imaging, radiotherapy, and surgical techniques,