Prediction method of 3D shape and model generating

		The surface patch extracted from the CT image (pink coloured dots) and the proximal femur model, predicted from the points of this patch show reasonable agreement.

Statistical shape models were implemented to initialise the segmentation of CT scans of the hip joint. Usually, model based methods do not allow detection of small-scale, individual features, which are not represented in the training set.

To overcome this difficulty, a refinement step was implemented, in which the bone surface was modelled by a mass-spring mesh that is deformed to match the CT image more accurately. The gradients of the image are analysed and patches of the organ boundary extracted and serve as attractors for the mass-spring mesh. Experiments demonstrated an accuracy of 1.20 mm.

An additional feature allows predicting the femoral head solely based on data of the distal femur. This option allows the application of our method even in cases of severe arthritis or in the presence of metal artefacts.