ABSTRACT
Background
Preprocedural planning with multislice computed tomography (MSCT) is an essential prerequisite for successful transcatheter aortic valve replacement (TAVR). However, preprocedural planning is time‐consuming and may contain clinically meaningful interobserver variability. Artificial intelligence (AI) based automated planning methods may provide anatomical MSCT measurements while reducing the time load required for the analysis. Previous studies on the validation of AI‐based TAVR planning methods excluded anatomical variants or lacked data on actual TAVR prosthesis size agreement. Mimics for TAVR (Materialise, Leuven, Belgium) is a new, AI‐driven TAVR planning platform based on patient‐specific 3D anatomical models and automated measurements through a web‐based user interface.
Aims
To validate the Mimics for TAVR planning tool in clinical practice.
Methods
Key planning assessments were performed on pre‐TAVR MSCT datasets using Mimics for TAVR and a commercially available semi‐automated software package. Agreement between the two planning methods was assessed using intraclass correlation coefficients and Bland‐Altman plots.
Results
A total of 126 consecutive patients (mean age 80 [74–85]; 56% male) who underwent MSCT as part of their TAVR work‐up were included. Mimics for TAVR derived measurements showcased high accuracy and excellent agreement for the area of the aortic annulus (mean error ± SD: −1.9 ± 3.3%, ICC: 0.976 [95% CI: 0.966–0.984]), left ventricular outflow (−2.7 ± 3.8%, ICC: 0.970 [0.957‐0.983]), sinotubular junction (−2.6 ± 4.2%, ICC: 0.953 [0.921–0.971]), coronary heights (LCA −3.5 ± 4.8%, ICC: 0.900 (0.859–0.928). RCA −3.9 ± 4.7%, ICC: 0.928 [0.898–0.949]) and membranous septum length (−5.0 ± 9.2%, ICC: 0.919 [0.880–0.940]). Mimics for TAVR assessment significantly reduced the time required for pre‐TAVR analysis (1.03 ± 0.15 min vs. 9.55 ± 3.16 min per case, respectively (p = 0.001).
Conclusion
The automated AI‐driven Mimics for TAVR planner is an accurate alternative to existing semiautomated TAVR planning platforms and may enhance overall workflow efficiency.
1. Introduction
Transcatheter aortic valve replacement (TAVR) is a minimally invasive alternative to surgical aortic valve replacement for patients with severe aortic stenosis. Preprocedural planning with multislice computed tomography (MSCT) is an essential prerequisite for a successful TAVR procedure and involves precise assessment of aortic root sizing, coronary obstruction risk, and access strategy. Semi‐automated software packages have simplified the CT analysis process, albeit it remains relatively time‐consuming, subject to interobserver variability, and prone to human error [1, 2, 3].
Fully automated TAVR planning platforms rely on artificial intelligence (AI) algorithms. Exploratory validation studies on fully automated TAVR planning software had high accuracy, but carried limitations as they either excluded anatomical variants (e.g., bicuspid valves, severely calcified valves, pre‐existing aortic valve prostheses) or lacked reports on the actual TAVR prosthesis size selection [4, 5, 6, 7, 8].
Materialise Mimics for Cardiac Interventions, in short “Mimics for TAVR” (Materialise, Leuven, Belgium) is a new, AI‐driven TAVR planner platform that generates patient‐specific 3D anatomical models and automated measurements through a web‐based user interface.
In this study, we tested the accuracy of aortic root measurements including transcatheter valve size selection by the Mimics for TAVR tool versus a commercially available semiautomated software package.
2. Methods
2.1. Study Design
In this prospective study, we included a total of 126 consecutive patients who underwent preprocedural MSCT screening for TAVR between June 2024 and November 2024.
Contrast‐enhanced electrocardiography (ECG)‐gated cardiac MSCT studies were performed on a dual source photon‐counting detector CT scanner (NAEOTOM ALPHA, Siemens Healthcare, Forchheim, Germany). A prospective ECG‐gated cardiac CT angiography (15%–45% of the R–R interval) was performed, including the heart and arterial tree. Imaging parameters included a tube voltage of 120 kV, an image quality setting of 34, and a collimation of 144 × 0.4 mm. The datasets were reconstructed using 0.4 mm slice thickness with 0.2 mm increments at 55 keV, using a Bv48 kernel and an iterative reconstruction strength of 4. The contrast administration protocol consisted of 45 mL of iodixanol (320 mg iodine/mL) injected at a flow rate of 3.0 mL/s. This was followed by an additional 25 mL of contrast medium at 2.5 mL/s and a subsequent 25 mL saline flush at the same flow rate.
MSCTs were uploaded in fully anonymized DICOM format to the web‐based Materialise Mimics Planner for Structural Heart Interventions platform (Mimics for TAVR) using Ambra, a picture archiving and communication system (PACS) direct link. Patient‐specific 3D anatomical models of the heart and arterial tree were created by an AI‐based automated anatomical segmentation tool and served as the basis for transcatheter valve sizing and overall procedure planning (Figure 1). In brief, the 3D modeling module relies on a 3D U‐Net8 for initialization and identification of the heart chambers and main vessels. The blood pool for both sides of the heart is then estimated, and a graph cut technique is employed to split the resulting segmentation mask into the desired structures. The landmark detection module relies on the same 3D U‐Net and was trained on a dataset composed of 449 CT scans and subsequently validated using 200 pre‐TAVR CT scans from the Montreal Heart Institute [9].
Figure 1.
Mimics planner user interface. [Color figure can be viewed at wileyonlinelibrary.com]
The patient‐specific case analysis was shared through the web‐based Mimics for TAVR platform. In the planner environment, additional conventional algorithms were used to create dedicated views, measurements and projections based on the 3D models and landmarks. For this validation study, the automated Mimics for TAVR data was utilized as the primary input; however, the software also permits manual modifications or the addition of supplementary measurements.
The accuracy of the automated Mimics for TAVR data was validated by comparison with pre‐TAVR measurements using the commercially available 3Mensio version 10 software (Pie Medical Imaging, Maastricht, Netherlands), which is the standard semi‐automated software package for TAVR planning at the Erasmus MC. 3Mensio measurements were independently performed by a team of three highly experienced physicians with experience in TAVR planning. For each parameter, the mean of three 3Mensio measurements was considered as the reference for validation of the Mimics for TAVR derived values. The interobserver variability among the three observers was calculated for all parameters that were analyzed by the 3Mensio and Mimics for TAVR packages.
A timer measured the total time a user spent obtaining a complete set of analyses with each tool, including all steps involving user input, interaction, and execution. In 3Mensio, this time accounted for user interaction during the measurement process, which included either manually performing measurements or confirming semiautomated measurement suggestions. Conversely, the initiation of the image upload to Materialise and the subsequent review of the AI‐derived measurements in Mimics for TAVR were timed.
We further separated complex and non‐complex anatomies for additional comparisons. Bicuspid phenotype, extensive aortic root calcification (Agatston score > 3000 men, > 1600 women), failing surgical/transcatheter aortic bio‐prosthesis and a small aortic annulus (i.e., < 430 mm2) were considered complex (Figure 2).
Figure 2.
Case examples of the complex cases cohort. (A) Case example displaying bicuspid aortic valve anatomy. (B) Case example displaying severe aortic valve calcification. (C) Case example displaying small aortic annulus dimensions. (D) Case example displaying a failing transcatheter bioprosthesis. [Color figure can be viewed at wileyonlinelibrary.com]
A separate analysis was conducted to assess the agreement in transcatheter valve size selection between the two TAVR planner methods. The transcatheter valve size was based on native aortic valve sizing by both methods and derived from the sizing charts in the instructions for use (IFU) of the respective TAVR platforms.
2.2. CT Image Assessment
Pre‐TAVR assessment involved the following measurements: Aortic annulus (minimum diameter, maximum diameter, average diameter, perimeter, area), left ventricular outflow tract (minimum diameter, maximum diameter, average diameter, perimeter, area) at 4 mm below the annular level, sino‐tubular junction (STJ) (minimum diameter, maximum diameter, average diameter, perimeter, area), sinus of Valsalva (SOV) (left coronary cusp (LCC), right coronary cusp (RCC), and noncoronary cusp (NCC) diameter), right and left coronary height (annulus to ostium), and the membranous septum (MS) length. A transcatheter valve size for the commercially available SAPIEN3 (Edwards Lifesciences, Irvine, USA), Evolut FX (Medtronic, Minneapolis, USA) and Navitor (Abbott, Illinois, USA) was proposed based on each set of analyses with Mimics for TAVR and 3Mensio.
2.3. Statistical Analysis
The distributions of continuous variables were tested for normality using the Kolomogrov‐Smirnov test. Continuous variables were reported as mean ± standard deviation or median (25th‐75th percentile) depending on distribution. Categorical variables were reported as number and percentage. To assess the correlation between the two TAVR planning methods, intraclass correlation coefficients (ICC) based on absolute agreement were calculated using a two‐way mixed‐effects model and reported with a 95% confidence interval (CI). Bland‐Altman plots were used to evaluate agreement further. For the 3Mensio measurement results, interobserver variability among the three observers was calculated. Mimics‐derived measurements for TAVR were automated and therefore were not subject to interobserver variability. All statistics were performed with SPSS software version 29.0 (SPSS, Chicago IL, USA).
3. Results
Study Population. We included 126 consecutive patients who underwent MSCT as part of their TAVR work‐up. Baseline characteristics are listed in Table 1. The mean age was 80 (74–85) years and 56% were male. Half of the overall population had complex anatomies: 30 patients had a severely calcified aortic valve, 12 had a bicuspid aortic valve, 7 had a failing surgical/transcatheter aortic bio‐prosthesis, and 14 had a small aortic annulus.
Table 1.
Baseline characteristics of the study population.
| Patient characteristics | (n = 126) |
|---|---|
| Age, years | 80 (74–85) |
| Male | 70 (56) |
| BMI, kg/m2 | 26.3 (24.0–29.2) |
| BSA, m2 | 1.8 (1.7–1.9) |
| Hypertension | 89 (71) |
| Hypercholesterolemia | 65 (52) |
| Stroke | 17 (14) |
| Diabetes mellitus | 44 (49) |
| CKD | 45 (36) |
| STS score | 5.5 (2.4–8.4) |
| Bicuspid aortic valve | 12 (10) |
| Severe calcification | 30 (24) |
| Small aortic annulus | 14 (11) |
| Failing bioprosthesis | 7 (5) |
| Echocardiography | |
| Left ventricular ejection fraction, % | 56 (43–67) |
| Aortic mean gradient, mmHg | 39 (17–49) |
| Aortic valve area, cm2 | 0.7 (0.5–1.3) |
| Aortic regurgitation | |
| None | 52 (41) |
| Mild | 43 (34) |
| Moderate | 25 (20) |
| Severe | 6 (5) |
Note: Values are median (interquartile range) or n (%).
Abbreviations: BMI = body mass index, BSA = body surface area, CKD = chronic kidney disease, STS = Society of Thoracic Surgeons.
3.1. Accuracy and Agreement of Pre‐TAVR Measurements
3.1.1. Total Cohort
Mimics for TAVR showed high accuracy as compared to 3Mensio. The mean difference for the aortic annulus area was −9.4 ± 18.7 mm2 (mean error −1.9 ± 3.3%). The mean difference for the aortic annulus perimeter was −1.2 ± 4.5 mm (mean error −1.5 ± 3.9%) (Table 2, Figure 3). Figure 4 shows a case example of measurement comparison. The mean difference for the LVOT area was −13.4 ± 28.3 mm2 (mean error −2.7 ± 3.8%). The mean difference for the LVOT perimeter was −1.5 ± 5.3 mm (mean error −1.8 ± 3.1%). Mean difference and error for additional measurements are listed in Table 2.
Table 2.
Accuracy of CT measurements (total cohort).
| Mimics‐ TAVR | 3Mensio structural heart | Mean difference | Mean error (%) | ICC | |
|---|---|---|---|---|---|
| Annulus | |||||
| Min D (mm) | 22.3 ± 2.5 | 22.9 ± 2.9 | −0.6 ± 1.6 | −2.6 ± 3.2 | 0.928 (0.895–0.949) |
| Max D (mm) | 27.8 ± 2.6 | 28.3 ± 2.6 | −0.5 ± 1.1 | −1.8 ± 4.1 | 0.944 (0.920–0.962) |
| Avg D (mm) | 25.0 ± 2.4 | 25.4 ± 2.5 | −0.4 ± 1.1 | −1.4 ± 2.9 | 0.946 (0.924–0.962) |
| Perimeter (mm) | 78.7 ± 10.3 | 79.9 ± 10.6 | −1.2 ± 4.5 | −1.5 ± 3.9 | 0.947 (0.925–0.963) |
| Area (mm2) | 480.5 ± 90.1 | 490.0 ± 91.3 | −9.4 ± 18.7 | −1.9 ± 3.3 | 0.976 (0.966–0.984) |
| LVOT | |||||
| Min D (mm) | 21.8 ± 2.8 | 22.1 ± 3.0 | −0.3 ± 1.4 | −1.1 ± 2.7 | 0.932 (0.903–0.952) |
| Max D (mm) | 29.1 ± 3.1 | 29.6 ± 3.2 | −0.5 ± 1.4 | −1.7 ± 4.0 | 0.949 (0.928–0.965) |
| Avg D (mm) | 24.9 ± 2.6 | 25.4 ± 2.9 | −0.6 ± 1.3 | −2.2 ± 3.8 | 0.940 (0.915–0.958) |
| Perimeter (mm) | 78.7 ± 10.0 | 80.1 ± 11.9 | −1.5 ± 5.3 | −1.8 ± 3.1 | 0.955 (0.923–0.973) |
| Area (mm2) | 476.5 ± 97.8 | 489.9 ± 96.4 | −13.4 ± 28.3 | −2.7 ± 3.8 | 0.970 (0.957–0.983) |
| STJ | |||||
| Min D (mm) | 24.9 ± 2.8 | 25.9 ± 3.0 | −1.0 ± 1.3 | −3.7 ± 3.9 | 0.948 (0.926–0.963) |
| Max D (mm) | 27.0 ± 3.1 | 27.9 ± 3.4 | −0.9 ± 1.5 | −3.0 ± 3.0 | 0.947 (0.925–0.963) |
| Avg D (mm) | 25.8 ± 3.0 | 26.7 ± 3.3 | −0,9 ± 1.4 | −3.4 ± 3.3 | 0.945 (0.922–0.962) |
| Perimeter (mm) | 100.2 ± 23.8 | 104.3 ± 25.4 | −4.1 ± 7.6 | −3.9 ± 4.1 | 0.965 (0.943–0.978) |
| Area (mm2) | 613.1 ± 132.5 | 629.8 ± 138.1 | −16.7 ± 28.8 | −2.6 ± 4.2 | 0.953 (0.921–0.971) |
| Sinus of Valsalva | |||||
| LCC (mm) | 32.9 ± 4.3 | 33.9 ± 5.2 | −1.0 ± 2.1 | −2.9 ± 4.7 | 0.947 (0.924–0.963) |
| RCC (mm) | 31.3 ± 3.9 | 32.5 ± 4.4 | −1.1 ± 1.7 | −3.5 ± 4.3 | 0.949 (0.928–0.965) |
| NCC (mm) | 32.8 ± 4.2 | 34.2 ± 4.5 | −1.4 ± 1.7 | −4.1 ± 3.4 | 0.922 (0.890–0.945) |
| Distance from annulus | |||||
| LCA ostium (mm) | 13.7 ± 3.5 | 14.2 ± 3.9 | −0.5 ± 1.6 | −3.5 ± 4.8 | 0.900 (0.859‐0.928) |
| RCA ostium (mm) | 16.0 ± 4.1 | 16.7 ± 4.4 | −0.7 ± 1.7 | −3.9 ± 4.7 | 0.928 (0.898–0.949) |
| MS (mm) | 5.7 ± 2.7 | 6.0 ± 2.6 | −0.3 ± 1.2 | −5.0 ± 9.2 | 0.919 (0.880–0.940) |
Note: Values are mean ± SD or ICC (95% CI).
Abbreviations: Avg D = average diameter, LCA = left coronary artery, LCC = left coronary cusp, LVOT = left ventricular outflow tract, Max D = maximum diameter, Min D = minimum diameter, MS = membranous septum, NCC = non coronary cusp, RCA = right coronary artery, RCC = right coronary cusp, SOV = sinus of Valsalva, STJ = sinotubular junction.
Figure 3.
Mean percent error for all measurements. Mean error (%) through Mimics for TAVR as compared to 3Mensio. The negative mean error (%) reflects underestimation through Mimics for TAVR as compared to 3Mensio. Avg D = average diameter, LCA = left coronary artery, LCC = left coronary cusp, LVOT = left ventricular outflow tract, Max D = maximum diameter, Min D = minimum diameter, MS = membranous septum, NCC = noncoronary cusp, SOV = sinus of Valsalva, STJ = sinotubular junction, RCA = right coronary artery, RCC = right coronary cusp, SD = standard deviation. [Color figure can be viewed at wileyonlinelibrary.com]
Figure 4.
Case example of aortic annulus measurement through both planning methods. (A) Aortic annulus measurement in Mimics for TAVR. (B) Aortic annulus measurement in 3Mensio. (C) Comparison of aortic annulus measurement results. Smaller Mimics for TAVR derived measurements as compared to 3Mensio derived measurements are reflected in a negative error. [Color figure can be viewed at wileyonlinelibrary.com]
Mimics for TAVR showed excellent agreement for the aortic annulus measurements (area: ICC: 0.976 (95% CI: 0.966–0.984); perimeter: ICC: 0.947 (0.925–0.963) and LVOT measurements (area: ICC 0.970 [0.957‐0.983]; perimeter: ICC: 0.955 [0.923–0.973]) (Figure 5). ICCs of additional measurements are listed in Table 2.
Figure 5.
Correlation and Bland‐Altman plots for annulus and LVOT measurements. (A and B) Correlation and Bland‐Altman plot for annulus area. (C and D) Correlation and Bland‐Altman plot for annular perimeter. (E and F) Correlation and Bland‐Altman plot for LVOT area. (G and H) Correlation and Bland‐Altman plot for LVOT perimeter. [Color figure can be viewed at wileyonlinelibrary.com]
3.1.2. Noncomplex Cases
The mean difference was −6.8 ± 19.0 mm2 (mean error −1.3 ± 3.9%) for the aortic annulus area and −1.1 ± 4.3 mm (mean error −1.3 ± 5.2%) for the aortic annulus perimeter. The mean difference was −11.7 ± 27.6 mm2 (mean error −2.3 ± 6.5%) for the LVOT area and −1.4 ± 5.2 mm (mean error −1.7 ± 5.3%) for the LVOT perimeter. Mean difference and error for other measurements are listed in Table S1. Mimics for TAVR showed excellent agreement for the aortic annulus measurements (area: ICC: 0.985 [0.976‐0.992]; perimeter: ICC: 0.962 [0.936‐0.977]) and LVOT measurements (area: ICC 0.979 [0.965‐0.987]; perimeter: ICC: 0.945 [0.915–0.968]). ICCs of additional measurements are listed in Table S1.
3.1.3. Complex Cases
The mean difference was −10.3 ± 18.0 mm2 (mean error −2.2 ± 4.1%) for the aortic annulus area and −1.3 ± 3.4 mm (mean error −1.7 ± 4.8%) for the aortic annulus perimeter. The mean difference was −14.3 ± 25.6 mm2 (mean error −3.1 ± 6.0%) for the LVOT area and −1.3 ± 5.2 mm (mean error −1.7 ± 6.4%) for the LVOT perimeter. Mean difference and error for other measurements are listed in Table S2. Mimics for TAVR showed excellent agreement for the aortic annulus measurements (area: ICC: 0.974 [0.962–0.986]; perimeter: ICC: 0.961 [0.940–0.975]) and LVOT measurements (area: ICC: 0.960 [0.934–0.976]; perimeter: ICC: 0.926 [0.896–0.947]). ICCs of additional measurements are listed in Table S2.
3.1.4. Interobserver Variability
The interobserver ICC among 3Mensio measurements for annulus area was 0.992 (95% CI: 0.988–0.996); the interobserver ICC for annulus perimeter was 0.989 (0.982–0.994). The interobserver ICC for LVOT area was 0.993 (0.989–0.997); the interobserver ICC for LVOT perimeter was 0.990 (0.984–0.994). Interobserver ICCs of additional measurements are listed in Table S3.
3.1.5. Transcatheter Valve Size Agreement
The transcatheter valve sizing per IFU sizing chart resulted in 90% sizing agreement for the Sapien3 (area‐based sizing), 87% for Navitor (perimeter‐based sizing) and 88% for Evolut (perimeter‐based sizing) (Figure 6). When using Mimics, a smaller prosthesis size would be selected for Sapien3, Navitor and Evolut, in 7%, 9%, and 9%, respectively and a larger prosthesis size in 3%, 4%, and 3%, respectively.
Figure 6.
Prosthesis size agreement. IFU‐based sizing yielded an agreement in 90% for SAPIEN (area‐based sizing), 87% for Navitor (perimeter‐based sizing) and 88% for Evolut (perimeter‐based sizing). [Color figure can be viewed at wileyonlinelibrary.com]
Time efficiency of Mimics for TAVR planner. The time spent on the overall TAVR CT planning process with Mimics for TAVR was significantly shorter than with 3Mensio (1.03 ± 0.15 min vs. 9.55 ± 3.16 min per case, respectively (p = 0.001).
4. Discussion
In this prospective, comparative study, we demonstrate that the AI‐driven Mimics for TAVR planner showed excellent accuracy and agreement, minimal interobserver variability and a faster work pace as compared to a contemporary commercially available semi‐automated software package.
Comprehensive MSCT‐based evaluation of the patient's anatomy is an essential prerequisite for a successful TAVR procedure. Contemporary TAVR planning software packages facilitate precise assessments of aortic root dimensions, coronary take‐off heights and arterial vasculature for accurate transcatheter valve sizing and strategy selection. However, pre‐TAVR CT evaluation requires a learning curve, clinical experience and systematic manual interference that is time‐consuming and subject to inter‐observer variability [1, 2, 3]. Deep learning algorithms allow for automated segmentation of cardiac structures and detection of anatomical landmarks and may curtail interobserver variability and enhance reproducibility. Examples of automated TAVR analysis software are HEARTNavigator and 4‐TAVR, which have been shown to enable fast, accurate and precise measurements [4, 6, 8]. However, the clinical validation of these platforms was limited by the exclusion of more complex anatomies (e.g., bicuspid valve morphology, heavily calcified valves or bioprosthetic valves).
A recent validation of the FORSSMANN algorithm, which accounted for these anatomical variants, revealed high accuracy and excellent agreement compared with 3Mensio [7]. However, the FORSSMAN validation study was a retrospective study and restricted to Chinese subjects, which may limit its generalizability to populations with a different ethnicity. Furthermore, it lacked data on transcatheter valve size selection.
Our prospective study demonstrated excellent correlations between the automated pre‐TAVR measurements and the standard practice of semi‐automated planning. The accuracy of measurements in our study was high compared to previous studies (e.g., 95% confidence intervals from the Bland‐Altman plot for the annulus area was ±37 mm2 vs. ±50 to ±100 mm2 in previous studies on semi‐automated tools) [3, 6, 7]. We also found excellent correlation with an ICC > 0.900 and a mean percentage error < 4% for most measurements
This validation study included non‐complex and complex anatomies (i.e., highly calcified valves, bicuspid valves, failing prostheses, and small anatomies), which adds to its robustness. We consistently demonstrated an overall excellent correlation for both noncomplex and complex cases. Importantly, the 90% agreement in transcatheter valve size selection between the automated and the semiautomated planning tools attests to the clinical validity.
Overall, we found higher ICC for area as compared to perimeter. Blanke et al. previously reported that Hounsfield unit based contour detection algorithms may, depending on the software platform, result in jagged contours, which may disproportionately affect the reported perimeter, whereas the area may be less affected [10]. This effect may be particularly pronounced in small or irregularly shaped annuli. Additionally, the Mimics planner quantifies annular dimensions using a polygonal shape, whereas 3Mensio relies on manually placed segmentation points connected via cubic spline interpolation. We hypothesize that the combination of differences in contour indication and the polygonal representation versus spline interpolation underlie the relatively lower ICC observed for perimeter.
The overall CT analysis time per case was 8 min shorter with the automated tool as compared to the semi‐automated tool, which is substantial and clinically relevant for contemporary clinical practice that is characterized by growing TAVR case volumes through increased adoption and expanding indications [11].
Finally, the web browser‐based platform provides real‐time access to the planning results, which may enhance real‐time clinical decision‐making during heart team discussions or TAVR procedures. Apart from conventional reporting, the user‐friendly interface allows the display of anatomical relationships in 2D and 3D and the ability to manually modify or add supplementary measurements upon demand.
4.1. Limitations
As this is a single‐center validation study in a tertiary care center, the findings may not be generalizable to other institutions with different patient populations. Further validation across multiple centers and CT scanner types is required to confirm broader applicability. MSCTs were reviewed by experienced imagers without the involvement of an independent core laboratory.
5. Conclusion
The automated AI‐driven Mimics for TAVR planner is an accurate alternative to existing semiautomated TAVR planning platforms and may enhance overall workflow efficiency.
Ethics Statement
All patients consented to the use of anonymous data for research purposes and proceedings were in accordance with the declaration of Helsinki. The study did not fall under the scope of the Medical Research Involving Human Subjects Act per Institutional Review Board.
Conflicts of Interest
Joost Daemen has received institutional grant/research support from Abbott Vascular, Boston Scientific, ACIST Medical, Medtronic, Microport, Pie Medical, and ReCor medical, and consultancy and speaker fees from Abbott, Abiomed, ACIST medical, Boston Scientific, Cardialysis BV, CardiacBooster, Kaminari Medical, ReCor Medical, PulseCath, Pie Medical, Siemens Health Care and Medtronic. Rutger Jan Nuis has received research grant support from Vifor Pharma and Merill, and consulting fees from Edwards Lifesciences, Abbott, Boston Scientific. Sophie Deckx is an employee of Materialise. Matthew Sherman is an employee of Materialise. Janelle Schrot is an employee of Materialise. Anna Kozak is an employee of Materialise. Alexander Hirsch received a research grant from GE Healthcare and consultancy/speaker fees from GE Healthcare and Bayer. He is advisory board member of Medis and was MRI corelab supervisor of Cardialysis BV. Nicolas M. Van Mieghem has received grant support/research contracts from Abbott Vascular, Boston Scientific, Medtronic, Edwards Lifesciences, Daiichi Sankyo, Astra Zeneca and PulseCath BV and has received consulting/speaker fees from Abbot Vascular, Boston Scientific Corporation, Medtronic, Daiichi Sankyo, PulseCath BV, JenaValve, and Amgen. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.
Supporting information
Supporting File
Data Availability Statement
The data that support the findings of this study are available from the corresponding author upon reasonable request.
References
- 1. Astudillo P., Mortier P., Bosmans J., et al., “Automatic Detection of the Aortic Annular Plane and Coronary Ostia From Multidetector Computed Tomography,” Journal of Interventional Cardiology 2020 (2020): 9843275. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Saitta S., Sturla F., Gorla R., et al., “A CT‐Based Deep Learning System for Automatic Assessment of Aortic Root Morphology for TAVI Planning,” Computers in Biology and Medicine 163 (2023): 107147. [DOI] [PubMed] [Google Scholar]
- 3. Horehledova B., Mihl C., Schwemmer C., et al., “Aortic Root Evaluation Prior to Transcatheter Aortic Valve Implantation‐Correlation of Manual and Semi‐Automatic Measurements,” PLoS One 13, no. 6 (2018): e0199732. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Toggweiler S., Wyler von Ballmoos M. C., Moccetti F., et al., “A Fully Automated Artificial Intelligence‐Driven Software for Planning of Transcatheter Aortic Valve Replacement,” Cardiovascular Revascularization Medicine 65 (2024): 25–31. [DOI] [PubMed] [Google Scholar]
- 5. Santalo‐Corcoy M., Corbin D., Tastet O., et al., “TAVI‐PREP: A Deep Learning‐Based Tool for Automated Measurements Extraction in TAVI Planning,” Diagnostics 13, no. 20 (2023): 3181. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Kočka V., Bártová L., Valošková N., et al., “Fully Automated Measurement of Aortic Root Anatomy Using Philips Heartnavigator Computed Tomography Software: Fast, Accurate, or Both?,” supplement, European Heart Journal Supplements 24, no. SB (2022): B36–B41. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Wang M., Niu G., Chen Y., et al., “Development and Validation of a Deep Learning‐Based Fully Automated Algorithm for Pre‐Tavr Ct Assessment of the Aortic Valvular Complex and Detection of Anatomical Risk Factors: A Retrospective, Multicentre Study,” EBioMedicine 96 (2023): 104794. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Meyer A., Kofler M., Montagner M., et al., “Reliability and Influence on Decision Making of Fully‐Automated Vs. Semi‐Automated Software Packages for Procedural Planning in TAVI,” Scientific Reports 10, no. 1 (2020): 10746. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Corbin D., Santaló‐Corcoy M., Tastet O., et al., “Validation Study of Two Artificial Intelligence‐Based Preplanning Methods for Transcatheter Aortic Valve Replacement Procedures,” Journal of the Society for Cardiovascular Angiography & Interventions 3, no. 3 Part B (2024): 101289. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Blanke P., Weir‐McCall J. R., Achenbach S., et al., “Computed Tomography Imaging in the Context of Transcatheter Aortic Valve Implantation (TAVI)/Transcatheter Aortic Valve Replacement (TAVR),” JACC: Cardiovascular Imaging 12, no. 1 (2019): 1–24. [DOI] [PubMed] [Google Scholar]
- 11. Généreux P., Leon M. B., Dar R. D., et al., “Predicting Treatment of Bioprosthetic Aortic Valve Failure in the United States: A Proposed Model,” Structural Heart 9 (2024): 100339. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Supporting File
Data Availability Statement
The data that support the findings of this study are available from the corresponding author upon reasonable request.