PURPOSE Novel imaging methods, such as ultra-fast Magnetic Resonance Imaging (MRI), enable high-resolution image acquisition in utero. As the assessment is currently performed qualitatively, clinicians and researchers lack a comprehensive reference to quantify developmental characteristics, and its variability in the population. We propose a method to build a spatio-temporal latent atlas from a single annotated example and a large number of non-annotated examples. The atlas captures the development of a cerebral structure in healthy fetuses during the 20th-30th gestational week. METHOD AND MATERIALS We perform a spatio-temporal group-wise segmentation of fetal brain structures given a single annotated example. The method is based on a spatio-temporal latent atlas capturing age-dependent characteristics in the training population which aids brain structure segmentation. The emerging atlas segments subcortical structures by integrating information across large number of subjects. It encodes the average development and its variability relevant for diagnosis. Furthermore, we re-estimate each subject's age by optimizing an energy function whose minimum is at the best age fit to the cohort. RESULTS Experiments show that our proposed spatio-temporal latent (ST) atlas outperforms an existing atlas (AVG) approach without age specificity, since it learns the time dependent shape of the structure during segmentation. Although the segmentation accuracy improves (Dice score ST/AVG atlas: 0.48/0.44), we expect further improvement by including a larger training sample, and more accurate non-rigid initial registration. CONCLUSION We propose a probabilistic spatio-temporal latent atlas for the segmentation of fetal brain structures during early development. From a single annotated example we learn an atlas and segmentations for a set of images. The benefits of the atlas are: 1. It serves as prior during segmentation of large numbers of structures that undergo development. 2. The atlas itself is informative regarding the developmental process. 3. The gestational age estimation explains shape variability by age shifts, which is relevant in a clinical context where the developmental process and possible pathological deviations have to be assessed. CLINICAL RELEVANCE/APPLICATION A spatio-temporal atlas provides a comprehensive quantitative reference that captures the characteristics and variability of the cerebral development in the population.
|State||Published - 28 Nov 2011|