Landmark-guided Hip Segmentation in 3D MR Images of a Large-Scale Cohort Study

摘要

Studying subtle prevalent degenerative findings of the hip in high resolution 3D MRI is of great interest. To reliably analyze femoral acetabular impingment~1 and further precursors of osteoarthritis, automatic hip bone segmentation becomes a mandatory prerequisite with respect to large-scale cohort data such as the German National Cohort (NAKO)2. Subsequently, accurate geometrical and structural properties can be derived and quantitative evaluation can be performed. Leveraging recent Deep Learning (DL) advancements a neural network architecture named MedPatchNet3 is extended, by providing additional anatomical landmarks for guidance. Thereby, accurate semantic hip bone segmentation is investigated.