Efficient deformable model with sparse shape composition prior on compromised right lung segmentation in CT

We developed an automated lung segmentation method, which uses deformable model with sparse shape composition prior for patients with compromised lung volumes with severe pathologies in CT. Fifteen thoracic computed tomography scans for patients with lung tumors were collected and reference lung ROIs in each scan was manually segmented to assess the performance of the method. First, sparse shape composition model is constructed using training dataset. Next, the deformable model with SSC prior will be initialized according to the rough segmented right lung ROI. Then, the right lung with compromised lung volumes is segmented using the robust deformable model. Energy terms from ROI edge potential and interior ROI region based potential are combined in this model for accurate and robust segmentation. The quantitative results of our segmentation method achieved mean dice score of (0.86, 0.97) with 95% CI, mean accuracy of (0.93, 0.98) with 95% CI, and mean relative error of (0.07, 0.17) with 95% CI. The qualitative and quantitative comparisons show that our proposed method can achieve better segmentation accuracy with less variance compared with a robust active shape model method (RASM). The proposed method will be useful in radiotherapy assessment in thoracic computed tomography and image analysis applications for lung nodule or lung cancer diagnosis.