Section: New Results

Discriminative Training of Deformable Contour Models

Paticipants: Haithem Boussaid, Iasonas Kokkinos and Nikos Paragios

Deformable Contour Models (DCMs) are a main workhorse for medical image analysis - but are not commonly studied from a machine learning perspective. In [21] , [20] we haved proposed an integrated machine learning and optimization framework to deploy DCMs in medical image analysis.

Our technical contributions are two-fold: firstly, we use an efficient decomposition-coordination algorithm to solve the optimization problems resulting from Loopy DCMs, by means of the Alternating Direction Method of Multipliers (ADMM); this yields substantially faster convergence than plain Dual Decomposition-based methods.

Secondly, we use structured prediction to exploit loss functions that better reflect the performance criteria used in medical image segmentation. By using the mean contour distance (MCD) as a structured loss during train- ing, we obtain clear test-time performance gains.

We demonstrate the merits of exact and efficient inference with rich, structured models in a large X-Ray image segmentation benchmark, where we obtain systematic improvements over the current state-of-the-art.