Inserting additional conditional atlas to improve registration accuracy during testing

Image registration and atlas construction are both critical tasks in medical image analysis, and they are closely interconnected. Previous atlas construction methods focus on the similarity between the constructed atlas and each individual image, but the resulting atlas often lacks clear anatomical structure, which makes alignment difficult. Most methods compute a single atlas for the entire image set or construct separate atlases for certain data subgroups. Using one or two conditional atlases as a registration bridge may still be insufficient for aligning images with large differences. Therefore, in this work, we explore the use of additional conditional atlases as a registration bridge to improve registration accuracy. We evaluated our method on the OASIS1 dataset, and the results show that using three conditional atlases achieved higher registration accuracy compared to using two conditional atlases.