Best Paper Award

Medical image segmentation is an essential and fundamental technique for automatic medical-image-analysis applications. The most popular approach of segmentation is energy minimization of the objective function based on the graph cut algorithm. While one-order energy function represents a relation of two pixels, high-order energy function can represent a relation among more than three pixels, such that a high-order energy function has recently been able to be optimized. The purpose of this paper, therefore, is to establish the efficient handling method of high-order graph-cut algorithms for medical image segmentation.

One of the significant advantages of this paper is to apply anatomical knowledge prior to the model construction and/or search criteria of minimization. In other words, a large number of evaluation variant sets and pixel relations, i.e. creeks, which are increased due to the high order function, are selected and reduced by the prior so that high quality segmentation is realized. The energy function in the sub-modular family, which is well studied by the authors, is also introduced to represent the objective function. A graph-cut based fast minimization algorithm of the energy function is proposed. The validity of the introduction of high-order energy functions for both efficiency and fastness is confirmed by the experimental results.

The usefulness of the proposed framework is shown by the three applications of pulmonary artery segmentation, coronary-artery lumen and plaque segmentation, and psoas major muscle segmentation from various CT images. In particular, pulmonary artery segmentation is well evaluated on not only an objective metric but also subject quality by clinicians. A medical imaging system based on this research is already in the market and its contribution is impressive for the improvement of both quality and effectiveness. Consequently, this paper is highly deserving of a Best Paper Award.