Machine Learning in Medical Imaging: A Comparative Review of Agglomerative and K-Means Clustering Techniques

Abstract

Brain tumor segmentation is a crucial aspect of medical image analysis, playing a significant role in accurate diagnosis, treatment planning, and patient monitoring. This study presents a comparative analysis of two widely used clustering algorithms—agglomerative clustering and K-means clustering—applied to a dataset of 100 MRI images of brain tumors. The objective is to evaluate the effectiveness of these unsupervised learning techniques in identifying tumor regions and to analyze their performance in terms of segmentation accuracy, adaptability to complex structures, and computational efficiency.

To ensure consistency and improve segmentation outcomes, the MRI images underwent preprocessing steps such as noise reduction, contrast enhancement, and normalization. The pre-processed images were then transformed into feature vectors using appropriate image descriptors. Agglomerative clustering, a hierarchical approach, iteratively merged similar data points to form clusters, making it well-suited for detecting tumors of irregular shapes. On the other hand, K-means clustering, which partitions data based on proximity to cluster centroids, demonstrated efficiency in segmenting tumor with more uniform and well-defined structures.

The experimental findings indicate that both algorithms successfully identified tumor regions, with distinct strengths and limitations. Agglomerative clustering proved effective in handling tumor with complex and arbitrary shapes, making it particularly suitable for segmenting irregular or asymmetrical tumor boundaries. However, it required higher computational resources due to its hierarchical nature. In contrast, K-means clustering exhibited faster processing times, making it more efficient for real-time applications, though it performed optimally when tumor regions were approximately spherical and of similar sizes. Visual inspections by domain experts validated the segmentation quality, highlighting that each algorithm had specific advantages depending on the tumor characteristics.

The insights derived from this study contribute to the advancement of medical image segmentation techniques by demonstrating the applicability of clustering algorithms in brain tumor detection. These findings can aid in the development of automated and more efficient segmentation methods, ultimately supporting medical professionals in improving diagnostic precision and patient care. Future research may explore hybrid approaches that combine clustering methods with deep learning techniques to enhance segmentation accuracy and robustness.