Year: 2025
Type: article
Abstract: Brain tumour segmentation and feature extraction from MRI are critical in aiding physicians in accurate diagnosis and treatment planning. Manual segmentation is labour-intensive, prone to observer variability, and can lead to inconsistencies in patient outcomes. To address these limitations, this study proposes a novel Merged Fuzzy C-Means and K-Means MFCMK-M algorithm for automated brain tumour segmentation to enhance accuracy, efficiency, and reliability. The proposed method was implemented in Python and evaluated against existing segmentation techniques. The Fuzzy C-Means algorithm initialises the K-Means clustering process, improving segmentation precision. Performance was assessed based on the Dice Similarity Coefficient (DSC) and computational efficiency, comparing the MFCMK-M with state-of-the-art models such as 3D U-Net, DeepMedic, U-Net, and 3D DenseUNet. The proposed MFCMK-M algorithm achieved a DSC of 0.85, surpassing 3D U-Net (0.81), and demonstrated superior segmentation performance. For smaller tumours (<10 cm3), the algorithm achieved a DSC of 0.86, and for larger tumours (>10 cm3), it achieved 0.88. In addition, the method achieved 95% accuracy, outperforming DeepMedic (90.7%), U-Net (91%), and 3D DenseUNet (91%). Computational efficiency was significantly improved, reducing the processing time to 50 s, making it 65% faster than Fuzzy C-Means and K-Means alone. Integrating Fuzzy C-Means and K-Means in the proposed algorithm provides an accurate and efficient approach to brain tumour segmentation, offering significant advancements over existing precision and computational speed methods.
Authors: Leonard Mac-Trevor, Kofi Sarpong Adu-Manu
Institution: University of Ghana
Language: English
Subfield: Neurology
Field: Neuroscience
Domain: Life Sciences
Open Access status: hybrid
OpenAlex ID: W7083696996