Segmentation of multiple heart cavities in wide-view fused 3D transesophageal echocardiograms

Three-dimensional transesophageal echocardiography (3D TEE) is an excellent modality for real-time visualization of the heart and monitoring of interventions. However, 3D TEE segmentation is still a challenging task due to the complex anatomy, typical ultrasound artifacts, and the limited field of view. To improve the segmentation accuracy of the left atrium we created wide-view TEE images by fusing several individual recordings. For five patients, six individual 3D TEE volumes were acquired (Philips X7-2t) by manipulating (e.g. rotating) the TEE probe head in the esophagus. CTA images were also acquired in these patients and segmented automatically by an atlas-based method, which served as ground truth (GT). The TEE volumes were manually registered by aligning them first with the CTA volume and afterwards with each other. The individual TEE sets were fused using a minimum intensity projection. Five cavities, left and right ventricle, left and right atrium and aorta (LV, LA, RV, LA, Ao) were segmented using a three stage segmentation scheme utilizing an Active Shape Model (ASM) and tissue probability maps estimated by a two class Gamma Mixture Model. The multi-cavity shape model was generated by Principal Component Analysis from a large database of segmented CTA images. The ASM stages involved a rigid transform fitting of the model, a shape updating stage for the whole heart model and a refinement stage for each individual cavity model. The Dice coefficients for the individual cavities between the TEE segmentations and the GT CTA segmentations were computed. We compared the quality of the segmentation results on the fused TEE sets with those using only the central single TEE view. There was a considerable improvement of the Dice coefficients for the fused data sets. The increase of the median Dice coefficients was 2, 5, 9, and 12 percent points for RV, LA, RA, and Ao respectively. These results show that image fusion significantly improves the segmentation of the c- vities close to the TEE probe head (LA, RA, and Ao).