AbstractIn this thesis, we investigate the potential for automation in the scoring process
of Psoriatic Arthritis x-rays. We focus on the identification of bones structures
through a latent space shape model that is driven by a Gaussian Process.
We describe a top to bottom approach to designing such a model that includes
the data collection and annotation process. We highlight the importance of cap-
turing and modelling uncertainties associated with having automated systems in
medical imaging. The main tool for this is noise models in a Bayesian setting.
The main mathematical contribution we make takes the form of a shape model
for which we perform an exact Bayesian marginalisation of the model parameters.
These parameters include the shape and the pose. We define a dependence struc-
ture that models the uncertainties present in a segmentation task. We show that
the Active Appearance Model of Cootes et al.  falls under our framework.
We believe that this is significant as previous work has only focused on the real
world performance of the models as opposed to the probabilistic interpretation.
Such an interpretation is important as it allows us to better understand the model uncertainties.
|Date of Award||13 May 2020|
|Supervisor||Gavin Shaddick (Supervisor), Tony Shardlow (Supervisor), Neill Campbell (Supervisor) & William Tillett (Supervisor)|
- PSORIATIC ARTHRITIS
- machine learning
- computer vision