Femur project: final steps

We have now reached the third and final step of our shape modelling project.

There are two major goals in this step:

1. Build a statistical femur model based on the data downloaded from the Sicas Medical Image Repository (SMIR).

2. Use this model to reconstruct a set of partial femur shapes.

Building a data-based femur model:

As you may have noticed, the femur shapes you downloaded from the SMIR repository are not in correspondence. Therefore, in order to build a statistical shape model from this data, you need to:

Fit the femur model you created in the first step of the project to each of the downloaded femurs. To do so, you can use the Iterative Closest Point (ICP) model fitting method we have seen this week.

Important: the provided femurs might strongly differ in size, making it difficult for the ICP method to obtain a good fit. Therefore, make sure to use the provided landmarks and build a posterior model before every registration in order to have a good size initialisation.

Once you have the model instance fitting as well as possible each provided femur mesh, use all fits to build a Principal Component Analysis (PCA) model of the femur. Make sure to store this model to file.

Complete the partial shapes:

To achieve the completion task, proceed as follows:

• Follow this link and download the partial femur shapes.

• Use the PCA model you have built from data to reconstruct the missing parts in the downloaded shapes. To do so, you can fit the model to the provided partial data (while paying attention in which direction you attribute candidate correspondences).
  Note here that you are not obliged to use the PCA model. Feel free to augment your built PCA model with any kernel you think would lead to better reconstructions.

Participating to the reconstruction competition:

Once you have performed the reconstruction for all the provided partial shapes, if you want, you can upload your meshes to the SMIR repository and evaluate how well you performed compared to your fellow learners.

To do so, follow the instructions on this page.