Abstract
Precise brain tumor segmentation in Magnetic Resonance Imaging (MRI) is challenging for healthcare providers and many researchers because of variability in size and boundaries. Conventional clustering and ensemble feature selection methods for complex tumor regions are characterized by uncertainty and can result in overfitting. To address these shortcomings, the proposed Hybrid FCM Ensemble framework integrates Fuzzy C-Means (FCM) clustering and wrapper filter feature selection with an ensemble classifier to address the segmentation challenges in MRI tumors. The multimodel magnetic resonance images are preprocessed using z-score normalization, Gaussian smoothing filtering and morphological techniques. The FCM wrapper filter method is used to reduce MRI heterogeneity, clustering and select unique tumor features. The three primary filtering criteria, such as solidity, area and major axis length are analyzed and set to standardized limits to ensure accurate segmentation results across multiple sizes of MRI images. These selected features are then processed by a weighted ensemble classifier and morphological refinement to enhance the segmentation process. The proposed framework is evaluated on the Multimodal Brain Tumor Segmentation Challenge (BraTS) 2018 dataset and the performance is evaluated with accuracy, dice, sensitivity and specificity metrics. The performance of 97.87 percent accuracy, 91.73 percent dice score with 92.38 percent sensitivity and 98.49 percent specificity indicates that the proposed method can effectively diagnose brain tumors compared with traditional methods.
Keywords: Brain Tumor, Clustering, Ensemble Techniques, Fuzzy Clustering, Segmentation, Wrapper-filter.