Enhanced Lung Segmentation in DICOM CT Images Using a Hybrid K-Means Clustering and Region Growing Technique

Abstract

Accurate lung segmentation is vital for effective lung cancer detection in CT scans. This paper presents a hybrid segmentation method that combines K-means clustering with region growing, enhanced by preprocessing techniques to improve overall accuracy. The preprocessing phase involves clipping pixel values to a specific range of Hounsfield Units  to minimize noise and normalizing the intensity values for consistent analysis. K-means clustering then segments the CT image into broad regions based on intensity, effectively identifying potential lung areas. Following this, region growing is employed to refine the segmentation. Seed points within a 64x64 pixel area are used to iteratively expand the segmented regions based on adaptive intensity thresholds determined from histogram analysis of the clustered regions. This combined approach leverages K-means for broad segmentation and region growing for fine-tuning, significantly enhancing the detection of lung cancerous lesions. Experimental results demonstrate the effectiveness of this method, showcasing its potential for improved automated lung cancer screening and diagnosis in clinical practice.