Fuzzy Clustering Based Noisy Image Segmentation of MRI/CT Scan Brain Tumor Images Using Different Distance Metrics as Similarity Measure

Medical images are captured using electronic devices such as CT scan and MRI machines in computer-aided diagnostics. However, these acquired CT scan/MRI images often exhibit limitations such as limited spatial resolution, poor contrast, noise, and non-uniform intensity changes due to environmental factors. Consequently, object boundaries become blurred and warped, leading to smudged object definitions. Fuzzy sets and fuzzy logic are effective in handling ambiguity and uncertainty, making them suitable for medical image analysis. Among the methods used for image segmentation in medical images, fuzzy clustering is preferred by researchers. The selection of a suitable distance metric is crucial for the clustering process, as the segmentation results in fuzzy clustering heavily rely on this choice. This study evaluates the performance of the fuzzy clustering process for segmenting noisy MRI/CT scan brain tumor images using various distance metrics, including Euclidean, Manhattan, Minkowski, kernel, kernel RBF, new distance, Canberra, Pearson, Chebyshev, and eisen cosine. To assess the performance, 2 digital and 10 CT scan/MRI brain tumor images are utilized. The effectiveness of the distance metrics for image segmentation is quantitatively evaluated using five metrics: partition entropy, partition coefficient, Fukuyama–Sugeno, Xie–Beni function, and Tanimoto index. Following a comprehensive qualitative and quantitative analysis of the segmentation results, it is observed that, in most cases, particularly in the presence of noisy images, the Manhattan distance metric outperforms other distance metrics for image segmentation while requiring the least execution time.