Patient Metadata-Constrained Shape Models for Cardiac Image Segmentation.

Patient metadata such as demographic information and cardio vascular disease CVD indicators are valuable data readily available in clinical practice. This information can be used to inform the construction of customized statistical shape models fitting the patient's unique characteristics. However, to the best of our knowledge, no studies have reported using these types of metadata in the construction of shape models for image segmentation. In this paper, we propose the use of a conditional model framework to include these patient metadata in the construction of a personalized shape model and evaluate its effect on image segmentation. Our validation on a dataset of 250 asymptomatic cardiac MR images shows an average segmentation improvement of 7﾿% and in some cases up﾿to 30﾿% over a conventional PCA-based framework. These results show the potential of our technique for improved shape analysis.