Ensemble Approach for Left Ventricle Segmentation

Chen Avni, Maya Herman

Abstract: This paper describes a novel algorithm for segmenting the Left Ventricle (LV) from cardiac MRI. Cardiac Magnetic Resonance Imaging (CMRI) is a non invasive medical imaging technique used in radiology to form a picture of the heart. The importance of precisely identifying the LV region in a cardiac MRI is crucial as it is the key in measuring LV volumes and determining cardiac function. Hence this paper describes a new algorithm based on an improved fuzzy c means clustering. To increase the segmentation accuracy min max ant system algorithm was used to initialize the cluster centers and principal components analysis algorithm used for time optimization. The rationale is to test fuzzy c means with ant colony optimization, a studied algorithm, in the task of LV segmentation. However, till now fuzzy c means with ant colony optimization was tested in radiology problems where small amount of data needed segmentation, contrary to this case in which doctors evaluate heart condition based on many cardiac MRI. Therefore this paper demonstrate how to use principal components analysis algorithm to tackle the problem of time performance. Algorithm was evaluated on real cardiac MRI with different heart conditions dataset. Results show that this approach improves the segmentation quality compared to other methods getting the highest BDP score (fraction of the correctly classified pixels) while reserving efficiency and accuracy.