Automatic Segmentation of Lung Parenchyma Using Fuzzy Clustering

Andr Fischer¹, Christian Oliver Ritter¹, Dietbert Hahn¹, Herbert Kstler¹

This work describes the ability of Fuzzy C-Means (FCM) clustering to accurately distinguish between pulmonary parenchyma, pulmonary vessels, the heart, and the surrounding tissue in dynamic contrast enhanced (DCE)-MRI. FCM clustering achieves this by clustering voxels with similar temporal signal courses together. A 3D DCE-MRI dataset was accordingly segmented and is presented in this work. This technique enables user independent automatic segmentation of the lung parenchyma necessary to quantify lung perfusion. Thereby, a subjective bias in data analysis as often present in manual parenchyma segmentation is lowered.