Bone segmentation in computed tomography images using a Hermite-based approach

Computed tomography images of the human bone system are essential for evaluation of abnormalities and disease detection. Structural and anatomical information can be assessed with computer tomography with the aim of performing diagnosis, planning and treatment evolution. Automatic segmentation can provide a fast, objective evaluation and quantification of the bone conditions. In this work, we propose a segmentation technique consisting of a region growing method implemented in the Hermite transform domain. The Hermite transform provides a powerful mathematical tool which is useful for extraction of the image features. These are obtained through a set of Hermite coefficients. A seed or a pre-segmentation is used to initialize the region growing approach and coefficients of the Hermite transform are posteriorly employed to grow the initial shape. We have used Hermite coefficients up to second order. Edge, gray level and zero crossing information obtained with the Hermite transform are configured for the growing criterion. Several computer tomography images were used for evaluation. Different metrics were employed for performance assessment and we have compared results of the proposed method against the manual segmentation. The obtained results demonstrate that the HT substantially improves the texture classification which is directly reflected into a better segmentation of the bone tissues. The region growing algorithm presents a better performance if it is applied to Hermite coefficients compared to the original method which is performed on the original image space.