Study design
The study protocol adhered to the Declaration of Helsinki and was approved by the Institutional Review Board. Written informed consent was obtained from all patients (ClinicalTrials.gov; NCT 02176616).
AMBER software algorithm for cardiac WT measurement
The myocardial WT calculation consists of three steps (Fig. 1). In the first step, the threshold for each tissue automatically calculated on the histogram of the cardiac CT (64 Channel, LightSpeed Volume CT, Philips, Brilliance 63, Amsterdam, The Netherlands), and the investigator draws the guidelines for the basis of the myocardial wall extraction by semi-automatically. The second stage extracted the myocardial wall region, and the third stage solved the Laplace equation to calculate a myocardial WT.
Myocardium boundary segmentation in cardiac CT
AMBER software was designed to automatically calculate the optimal endocardial boundary based on the edge protruding through the Sobel filter after removing noise with the median filter (Fig. 2a). While manually drawing the boundary line in the CT image, the multi-Otsu threshold algorithm [5] is applied in parallel to calculate the HU threshold for the density of each tissue. We assumed that cardiac CT will focus on the heart, and there will be four tissues with the highest frequency in the histogram of the HU (bone and blood pool, muscle tissue, fat, and air). Therefore, the multi-Otsu algorithm computes the histogram of the HU and the three thresholds that classify the four groups. While checking the results of the screen, the investigator selects one of the four tissue groups for detailed adjustment. For computational efficiency, all tissues of CT images were divided into 4 groups to calculate the automatically performed threshold by an independent thread running in parallel.
Extraction of myocardium wall
Extraction of cardiac wall was performed from the heart boundary line. From the boundary of endo- and epicardial surfaces obtained by the previous step, cross-overlaps with the HU of the myocardial tissue were extracted to the myocardial wall. In this process, two approaches were used for atrial WT and ventricular WT (Fig. 1b). For atrial WT measurement, cardiac wall was extracted by overlapping with the HU threshold of the myocardium while gradually expanding (\({\text{expanding coeff}} = \hbox{max} \left( {2/{\text{Pixel Spacing}}, 8} \right)\)) through morphological dilation from the endocardial boundary line (Fig. 2b). Since it progresses from the endocardium to the epicardium, the initial vector fields state for Laplace solve was determined to be 10,000 for the endocardium, 0 for the epicardium, and 5000 for the crossing area. In contrast, for ventricular WT measurement, the ventricular wall was defined from the epicardial boundary line to the endocardial surface because the ventricular endocardial surface is highly irregular due to papillary muscle and trabeculations. Since it is calculated in the opposite direction to the LA, the initial state for the Laplace solve was determined to be 10,000 at the epicardium, 0 at the endocardium, and 5000 at the intersection.
Laplace equation to solve wall thickness
The Laplace equation was applied to measure accurate cardiac WT in 3D space (Fig. 1c). The myocardial region was assumed to be topologically spherical. The orthogonal projection distance of all specific points P → P′ did not match P′ → P (Fig. 2c). Therefore, curves were connected to satisfy mutuality between the two points P and P′. The myocardial region was represented as the area for solving the differential equation. A vector field was formed containing a tangent vector along the field line connecting the initial vector field values set based on the boundary between the endocardium and epicardium. The Laplace equation stops when E = 10−5 or 400 iterations are met to generate an equipotential vector field. Finally, a differential equation for thickness calculation was performed by applying the Euler method of the vector field at all voxel points of the epicardium or endocardium (dt = 0.01).
Validation of AMBER in 3D phantom model
A 3D-phantom model was designed using FreeCAD (0.18.4) and produced with a 3D printer (output equipment: Veltz-600H, material: ABS-like, stacking thickness: 0.1 mm) (Fig. 3a). The 3D-phantom model was designed to reflect the thickness of the atrium and ventricle. Two half dome regions were divided into sections with 4 different WTs each and configured with different thicknesses accounting for bidirectional measurement from inside to outside and outside to inside. Contrast enhancement is evenly distributed along the inner surface of the atrium, while ventricles have various protruding structures along the inner surface. Therefore, it was derived to measure each WT based on a uniform surface. After acquiring CT images of the 3D-phantom model (Fig. 3b), AMBER-measured 3D-WT and the computer-aided design (CAD)-generated real WT were compared. The AMBER-WT maps were analyzed except for borders where thicknesses crossed each other because these are difficult to compare between the equipotential distance of the Laplace equation and orthogonal projection distance (Fig. 3c).
Validation of AMBER-measured atrial WT
Since it is impossible to measure real atrial WT in a living human atrium, LAWT measured by AMBER in 120 patients with AF was compared with regional LAWT in 12 previously published studies (Fig. 4) [6,7,8,9,10,11,12,13,14,15,16,17]. The LAWT was measured using the Laplace equation and Euler method described above using a 3D-AMBER map composed of about 90,000 nodes. To measure by region, the LA mesh was created using the marching cube algorithm for the AMBER segmentation information. LA was divided into 8 regions based on criteria that were previously used in CT imaging [18]. Using custom made software (CUVIA, Model: SH01, ver. 2.0; Laonmed Inc., Seoul, Korea), the dividing lines of the regions were displayed on the surface of the LA mesh [18, 19]. Based on these lines, a search algorithm was used to assign the number of each LA section to each node, and a quantitative thickness result corresponding to each region was exported (Fig. 4a–c). In the previous research on human LAWT, 516 studies were found based on the search keywords “human, atrium, and wall thickness,” and the data were collected by selecting 12 papers that presented human LAWTs by region. If the category by region of LA was different from this study, the location of the region presented in the paper was evaluated and the LAWT value of the region most similar was selected. The overall distribution of regional LAWT presented in a total of 12 studies and 120 AF patients in this study was compared (Fig. 4d).
Validation of AMBER-measured ventricular WT
The robustness of the new software was demonstrated by comparing AMBER-measured LVWT with WT measured by echocardiogram. The LVWTs in 16 standard segments [20] were evaluated and compared in 10 patients with both cardiac CT images and standard echocardiographic images (Fig. 5a). Echocardiographic parameters were obtained according to the American Society of Echocardiography guidelines [21]. For AMBER-measured LVWT analysis, a regional LV mesh model was generated by fitting CT images of each patient to 16 standard segments of echocardiogram at the QRS gating state. The output of AMBER-measured LVWT for 16 segments was compared with echocardiographically measured LVWTs of each segment.
Feasibility test for AMBER-WT map in clinical AF catheter ablation
We tried to verify clinical feasibility of AMBER-measured WT map during AF catheter ablation procedure for a patient. Electrophysiological mapping and radiofrequency catheter ablation have been described previously. Briefly, an open irrigated-tip catheter (Smart-touch [Johnson & Johnson Inc., Diamond Bar, CA, USA]; 30–35 W, 47 °C, contact force > 10–20 g) was used to deliver radiofrequency energy for the ablation under 3D electroanatomic mapping (CARTO3, Johnson & Johnson Inc. USA.) merged with 3D spiral CT. During circumferential pulmonary vein isolation (CPVI), the LAWT color map displayed in real-time on the 3D-mapping screen to confirm whether radiofrequency energy titration was possible by titrating the ablation index [22] depending on the local LAWT of the corresponding ablation site. After CPVI, bidirectional block was confirmed. The procedure was completed when there was no immediate recurrence of AF after cardioversion with an isoproterenol infusion (5–10 μg/min depending on ß-blocker use, target sinus heart rate 120 bpm).
Statistical analysis
Because the trial was conducted as a pilot study, the sample size was driven by the computation time and feasibility of recruiting patients. Patient enrollment was open for 6 months for other clinical trials of computational modeling (CUVIA-AF2, clinical trial.gov. NCT 02558699). Results are expressed as mean values ± standard deviation for continuous variables, and absolute number and percentages for categorical variables. Continuous variables were compared using Student’s t test, and categorical variables were compared using either Chi-square test or Fisher’s exact test as appropriate. Comparisons of AMBER-measured WT and real WT or echocardiographically measured WT were evaluated by linear correlation methods. A p value of < 0.05 was considered statistically significant. All statistical analyses were performed using SPSS version 18.0 software (SPSS Inc., Chicago, IL, USA) and the R package (3.1.0, R Foundation for Statistical Computing, Boston, Massachusetts, United States).