Unsupervised Lung Tumor Segmentation in CT Images Using K-Means and Hierarchical Clustering: A Comparative Analysis Toward Early Detection

Background: Lung cancer is among a top cause of cancer-related fatalities worldwide, and timely, accurate detection is essential to improving patient survival rates. Examining CT scans manually takes significant time and is susceptible to observer variation, creating the need for automated detection systems. Methods Used: This study presents an unsupervised segmentation framework for lung tumours in CT images using K-means and Hierarchical Clustering. The process involves grayscale conversion, noise filtering, contrast enhancement, clustering-based segmentation, and morphological post-processing. Results Achieved: Performance indicators such as the Dice Similarity Coefficient (DSC) and the Jaccard Index, tumor area, perimeter, and circularity show that hierarchical clustering provides more accurate and morphologically consistent results, while K-means is computationally faster but less precise. Concluding Remarks: The findings support the use of unsupervised clustering as an effective annotation-free approach for tumor segmentation and pave the way for future research into hybrid and deep learning-based models that can enhance segmentation accuracy and clinical applicability.