Detection of slices including a ground-glass opacity nodule in CT volume data with semi-supervised learning

The features of GGO nodules need to be obtained such as volume, mean, variance of Ground-Glass Opacity Nodules by boundaries of GGO nodules to judge malignant or benign of lung tumors. However, radiologists need to look for the slices including the GGO nodule in CT volume data. It is time-consuming. This paper proposes a semi-supervised learning method based on the label propagation. First, a GGO nodule was labeled in one slice. Secondly, similarities were found by comparing with the labeled GGO nodule using the values of pixels. Finally, the GGO nodule of the other slices was labeled by iteration. Experimental results showed that the approach of this paper can find slices including the GGO nodule. The approach is better than the nearest neighbor algorithm in performance.