Structure-aware preserving projections with applications to medical image clustering

Keyang Yu, Yike Zhu, Xuesong Yin, Ting Shu, Yigang Wang, Enliang Hu

Abstract: Highlights•Our algorithm exploits a powerful combination of subspace clustering and nearest neighbor graph construction to effectively capture the local structure of the data. Hence, the proposed SAPP introduces the grouping effect of the representation into projection learning.•SAPP is able to preserve the global structure and to capture the complex relationship of data points. In the embedded space, SAPP tends to segment highly correlated data points into a group.•The proposed algorithm can simultaneously exploit global and local structures to capture the accurate affinity between data points. Therefore, the learned projection has more discriminative power.•Extensive experimental results on medical image datasets show that the clustering accuracy of the proposed algorithm is superior to that of other state-of-the-art methods, indicating the effectiveness of our algorithm.