Meeting Banner
Abstract #1682

Whole Body Muscle Classification Using Multiple Prototype Voting

Anette Karlsson1, 2, Johannes Rosander3, Joakim Tallberg2, Thobias Romu1, 2, Magnus Borga1, 2, Olof Dahlqvist Leinhard, 24

1Department of Biomedical Engineering (IMT), Linkping University, Linkping, Sweden; 2Center for Medical Image Science and Visualization (CMIV), Linkping University, Linkping, Sweden; 3Advanced MR Analytics (AMRA) AB, Linkping, Sweden; 4Department of Medical and Health Sciences (IMH), Linkping University, Linkping, Sweden

Fat and water separated MRI enables non-invasive quantification of volume and fat infiltration in muscles. Manual segmentation of muscles is extremely time consuming why automatic alternatives are needed. . We have developed an infrastructure that enables a robust platform for non-rigid whole body registration where manual classifications of an anatomical structure in an image volume (prototype) may be automatically transferred to a new patient volume. The purpose of this work was to evaluate if using such multiple prototype voting procedure provides a robust automatic muscle classification. The result showed satisfying robustness in all 10 subjects using multiple prototype voting.

Abstract PDF Poster PDF

Keywords

abdominal able adaptive agree agreed agrees analytics anatomical another arms automatic averaging back best biomedical body border canvas chose classification classifications classified classify classifying clear combined consists contribute corresponds cross determine dimensional elastic enables equal estimation evaluate every exact females finding finds five front future generally gradient great ground half hand handle health highest humans included includes index infiltration infrastructure instead inverse lead leakage link local looking lowest mainly make males manual manually meant measure medical medicine minute muscle muscles must nine normalized novel occurred onto optimal overestimation performance ping post priors probability problem problems procedure processing prototype prototypes quantification quantitative reconstruction reduce registered registering registration rehabilitation remove repetition resolution rigid risk robust scale science sciences segmentation segmented sensitive separated separation seven several shape side similarity sports stands steps structural structure subjects sufficient systems target threshold tissue together transferred truth typical underestimation upper variable variation volume voting water whole years yielding