Click on image to view larger version.

Figure 1 In vitro left ventricular (LV) phantom experiments. After filling a hollow plastic model of the heart (A) with a dilute solution of gadolinium, magnetic resonance imaging (MRI) scans were performed. The aorta and LV were manually segmented to generate a three-dimensional (3D) surface reconstruction (B). (C) After electroanatomic mapping of this phantom, the locations of the endoluminal aortic (blue points) and endocavitary LV (white points) points were registered with the 3D MRI. Strategies to register these MRI and magnetic electroanatomical mapping (MEAM) datasets were then evaluated. (D) Image registration was based solely upon registration of the MEAM LV points with the MRI-based LV surface. After a coordinate transformation, each MEAM LV point was sequentially incorporated into the registration process (horizontal axis). The level of accuracy of the registration process (vertical axis) was defined as the mean distance of either: 1) the complete MEAM LV point dataset to the MRI-based LV surface (red lines), or 2) the complete MEAM aorta point dataset to the MRI-based aorta surface (blue lines). Five simulations were performed with each of three separate dataset acquisitions; each line represents an average of each set of five simulations. This reveals that the MEAM LV point to MRI surface distance (red lines) can be minimized to 1 mm error, after only 30 points. However, the high level of inaccuracy in the MEAM aorta point to MRI surface distance (blue lines) demonstrates that the registration was simply a local minimal solution. Videlicet, in the registered image (right), the LV points appear to be well-aligned, but the misalignment of the aorta points indicates that this is an inaccurate solution—apparently as a result of rotation about the LV long axis. (E) Image integration was based upon first registering all of the MEAM aorta points with the MRI surface, followed by sequential incorporation of each MEAM LV point into the registrations process (horizontal axis). The level of accuracy of the registration process (vertical axis) was again defined by the mean distance of the complete MEAM LV point dataset to the MRI-based LV surface (red lines). The results show that after first registering the aorta, the MEAM LV point to MRI surface distance (red lines) can be minimized to 1 mm error after incorporating the first three LV points into the registration process.