Akademik Veri Yönetim Sistemi
Signal, Image and Video Processing, cilt.19, sa.11, 2025 (SCI-Expanded, Scopus)
Anomaly classification in medical imaging has emerged as a prominent area of research in recent years. Detecting cancers from brain MRI (Magnetic Resonance Imaging) scans is a critical challenge in medical imaging, significantly contributing to the diagnosis and treatment of brain malignancies. This procedure entails analyzing MRI pictures to detect and pinpoint abnormal growths or tumors in the brain. Developing dependable tumor detection tools is essential for the early diagnosis, treatment, and monitoring of brain tumors. Analysis of the research reveals a preference for convolutional neural network (CNN) based methods to enhance tumor detection performance in brain MR images. This study utilizes ResNet101 and DarkNet53 pre-trained CNN models to extract features from MR images. Heuristic techniques were employed to identify relevant traits. The chosen characteristics were trained and assessed using support vector machines. During the feature selection phase, binary particle swarm optimization (bPSO), binary firefly algorithm (bFA), binary ant colony optimization (bACO), binary harmony search (bHS), and binary path finder algorithm (bPFA) were employed. Furthermore, a new hybrid sequential binary optimization algorithm named bPSOFA was introduced utilizing techniques that yield effective outcomes. The proposed methods were evaluated using the Brain Tumor MRI Dataset (7023 images) and Brain MRI Images for Brain Tumor Detection (253 images) datasets. Experimental results show that the proposed bPSOFA based DarkNet53 model achieves 100% classification accuracy on the Brain MRI Images for Brain Tumor Detection dataset. Additionally, the proposed bACO based DarkNet53 model achieves an accuracy of 96.41% on the Brain Tumor MRI Dataset, outperforming several state-of-the-art approaches.