Myocardium segmentation improvement with anisotropic anomalous diffusion filter applied to cardiac magnetic resonance imaging

Cardiologic magnetic resonance imaging (MRI) has recently been improved by faster acquisition and higher resolution hardware. Commercially available MRI equipment is able to capture contrast agents with the needed time and space definition to map myocardial viability. MRI myocardial imaging has an emerging role in cardiology studies, and it has experienced a crescent relevance in clinical investigations. Although MRI has potential for clinical investigation and application, an efficient digital filter is needed in order to allow robust myocardial segmentation. This paper proposes anisotropic anomalous diffusion (AAD) filtering to reduce noise levels while preserving myocardial traits. The proposed AAD filter follows the porous media equation consistent with inhomogenous complex media, and thus appropriate to model biological systems. In this study, the porous media equation together with gradient driven diffusion has been applied to digital image smoothing. Eleven MRI T1 weighted cardiology images were used hereby to evaluate both the AAD and classical Gaussian filter in a segmentation pipeline. in order to study the filtering application in a automatic segmentation algorithm (Geodesic Active Contour). The myocardial area, i.e. epicardic and endocardic border, was delineated with both the AAD and Gaussian filter. We calculated the root mean square error, when compared to the manual traces, to measure automatic segmentation quality. The AAD filter show a significant segmentation accuracy enhancement (p < 0.001), while no significant difference was found between the AAD filtered and manually segmented images. The findings suggest that AAD filtered image segmentations have similar reliability to manual segmentation.