Ground-Glass-Opacity Nodule Detection and Segmentation Based on Dot Filter and Gaussian Mixture Model Hidden Markov Random Field

Aiming at solving the problem that ground-glass-opacity (GGO) nodules cannot be detected directly by "dot" filter, a method based on vessel elimination and "dot" filter was used to detect them. Because of spatial properties consideration, it is proposed to use hidden Markov random field based on Gaussian Mixture Model (GMM-HMRF) and Expectation-Maximization (EM) algorithm to segment GGO nodule in this paper. Experiments were performed at 87 scans of CT images of lung alveoli. (containing 36 GGO nodule). The sensitivity of the detection method was 83.3%. The time cost was 1.2 minutes per scan. The accuracy of the segmentation method was 80.9%. This new methods are superior to the existing methods on the required time and sensitivity.