Lung Segmentation and Tumor Detection from CT Thorax Volumes of FDG PET-CT Scans by Template Registration and Incorporation of Functional Information

Automatic segmentation and detection of lungs find tumors in FDG PET-CT images is potentially beneficial in the diagnosis and staging of patients with non-small cell lung cancer (NSCLC). However, simultaneous lung segmentation and tumor detection is not a trivial task, particularly due to noise in the datasets, proximity of the lung lesion to the mediastinum find chest wall in certain instances, and disease involvement of non-enlarged lymph nodes. We propose a novel, automated segmentation method for PET-CT thoracic volumes that is able to simultaneously segment lung structures and lung tumors. In the initial lung segmentation, histogram analysis is used to set a fixed threshold for the initial region growing. Then, automatic elastic registration is used to build a probabilistic lung atlas from healthy lung CT volumes, and to align the template to the patient image volumes. In the current registration-based segmentation methods, the direct use of the template to define the lungs may introduce small errors in lung segmentation due to misalignments from inter-subject registration. In our method, the differences between the lung template and the initial lung segmentation are used as tumor candidates. By incorporating functional information from PET images, tumors can then be automatically localized. Finally, in the lung refinement, full lung region definition can be achieved by integrating the tumor and the initial lung. Our initial experiments on clinical datasets of patients with NSCLC show that the method can segment tumors and the lungs automatically from PET-CT data.