Automatic Lung Ventilation Estimation Using Supervoxel Multi-Feature Fusion Combined with CT Image Registration

Purpose: In order to reduce the influence of respiratory motion artifacts on lung ventilation estimation method, a lung ventilation estimation method based on multi-feature fusion of supervoxels (MFFS) and CT image registration is proposed. Methods: In MFFS, the Simple Linear Iterative Clustering (SLIC) algorithm is used in the initial layer of supervoxel clustering to solve the problem of large errors produced voxel level pulmonary ventilation estimation method. Combined with geometric feature and gradient feature of CT lung images, an objective function of local weighted multi-feature fusion that contains clustering prior knowledge is designed to optimize and subdivide initial supervoxel. The inspiratory and expiratory peak images are registered using deformable image registration (DIR) method to obtain the image deformation field, which is then used to calculate the variations of CT values in the supervoxel region for obtaining ventilation volume of supervoxel region. The expiratory peak image is represented by multiple layers of supervoxels, and the final CT ventilation estimation image (CTVI) is acquired throuogh averaging all supervoxel layers, which can effectively enhance the robustness of the MFFS method. Results: Experimental results demonstrate that the proposed MFFS method combined with CT image registration can obtain an average Spearman coefficient of 0.43 between the CTVIs and public VAMPIRE dataset. Conclusion: In contrast to classical lung ventilation estimation methods, MFFS method completely considers the contribution of multiple features of supervoxel regions in lung images to clustering centres, resulting in more regular lung tissue shapes.