||We have developed a registration pipeline based on an assumption that bone is relatively rigid and has low inter-subject variation, lean tissue is less rigid and has higher inter-individual variation, and adipose tissue is highly elastic and has high inter-subject variation. The aligned whole-body volumes are then compared by a point-wise statistical analysis, where we, for example, compute correlations to non-imaging parameters, anomaly detection and group-wise comparisons.
The current work on the technical development of the image registration includes a non-parametric registration method and a machine-learning approach for pre-segmentation of abdominal organs. Also, the performed statistical tests are not independent, and therefore, we develop methods for correction of multiple tests, including permutation based methods.
The methodology is currently applied in diabetes, oncology, and obesity. We are aiming to further develop the methodology for genetic analyses.
In this seminar, I will discuss our current technical solution and potential directions of future work. I will also present some medical applications of the methodology.