Computer-Aided Diagnosis relies on accurate tumor volume and heterogeneity assessment through CT-scans. Precise lesion segmentation is essential for patient diagnosis, therefore the development of automatic tools for lesion segmentation is needed. To improve lung nodule segmentation performance, lung segmentation masks serve as valuable priors, narrowing the focus to lung regions. Current methods suffer from the exclusion of pathological areas, especially in oncology patients, since tumor tissue differ in voxel density from other structures in the lung. Consequently, ensuring accurate lung segmentation encompassing all lesions is crucial. We developed a method based on supervoxels to fully segment the lung while encompassing nodules if present using a propagation algorithm based on geometrical properties. We compared our method to a morphology based method and neural networks trained to segment the lungs. Our method had the best performance in the inclusion of lung lesions, while retaining an adequate level of precision.