This paper presents a segmentation process of the myocardium, endocardium and epicardium surfaces of the heart from 3D SPECT images to compute a heterogeneity index. This index represents the distribution of the activity in the myocardium. Because of the low resolution of SPECT images, the thickness of the myocardium is 1 to 4 voxels and a sub-voxel accuracy is therefore needed. The segmentation process is based on the minimization of an energy by dynamic programming after a coarse segmentation to define the center surface of the myocardium. A Gaussian mixture is fitted on the data to ensure subvoxel accuracy. The heterogeneity index is compared to the reference index on 58 SPECT images and the segmentation is visually validated on 300 SPECT images by a clinician.