Automated Recognition of Abnormal Left Ventricle Wall Motion

Introduction: Currently, diagnosis of hypokinetic left ventricle (LV) wall motion is generally based on visual inspection of cardiac magnetic resonance cine images. Studies have shown that this method may be inaccurate, time consuming and suffer from high inter-observer variability [1]. Therefore we propose an algorithm for automated detection of abnormal wall motion that includes inter-subject image normalization and a pattern recognition technique. Materials and Methods: Imaging data (N=17) were acquired on a 1.5T MRI (GE CV/i Excite) using IR-SSFP short axis (SA) cine sequence; 12/17 patients had regional abnormal wall motion by expert visual assessment. For each of the 17 subjects, a single basal slice was selected for this preliminary study. The recognition algorithm consists of three stages. First, manual input was required to define the LV centre and a reference point on the epicardium. The LV images were normalized by spatially mapping the pixels from Cartesian to polar coordinates, intensity normalization, and labeling standard segments (Fig.1). The essential feature of the cardiac images is the radial wall motion; therefore the analysis is simplified by mapping the pixel intensities from Cartesian (x, y) to polar (r, θ) coordinates. Also, the mapping normalizes the size and shape of the myocardium by equalizing the length of each radial line from the LV centre to the epicardium. Voxel intensities are normalized by setting each to σ μ / ) ( − x , where x is the original intensity, and μ and σ are respectively the intensity mean and the standard deviation (see Fig.1 b & c.)