An improved supervoxel 3D region growing method based on PET/CT multimodal data for segmentation and reconstruction of GGNs

Among the various types of lung nodules, ground glass nodules (GGNs) are difficult to segment accurately due to complex morphological characteristics. Moreover, GGNs are associated with a higher malignancy probability. Three-dimensional (3D) segmentation and reconstruction techniques can help physicians intuitively elucidate the relationship between lung nodules and their surrounding tissues. We propose an improved supervoxel 3D region growing approach based on positron emission tomography/computed tomography (PET/CT) multimodal data for the segmentation and reconstruction of GGNs. First, the seed point is automatically located with PET information and a 3D mask is generated. Then, a fuzzy connectivity (FC) map is generated based on the 3D mask, and an improved supervoxel 3D region growing is utilized on a fuzzy connectivity map under the constraints of the 3D mask. Finally, 3D GGNs segmentation and reconstruction results are obtained. Qualitative and quantitative comparisons between our proposed method and other region growing methods shows great superiority of our proposed method, with the Jaccard similarity coefficient between our proposed method and physician manual segmentation reaching 95.61%; the average processing time is 16.38 s. Experimental results show that our proposed supervoxel-based 3D region growing method is very promising for assisting physicians in diagnosis.