Comprehensive Summary
This study examines the efficacy of an artificial intelligence based software platform in analyzing the key measurements of computed tomography (CT) scans for transcatheter aortic valve replacement (TAVR) procedures. A data set of 825 CTs of patients with severe aortic stenosis were used to train the AI algorithm and 90 CTs were used to tune the software. The algorithm was validated with 100 CTs of TAVR patients that were also analyzed by experienced operators with software packages used for CT analysis currently. The measurements from the operators were averaged and compared to those made by the AI. These measurements were very similar, with a correlation coefficients of 0.97 for both the aortic perimeter and aortic area, 0.95 for the ascending aorta measured 5 cm above the annular plane, and 0.85 for sinotubular junction diameter. The researchers acknowledged the need for further validation with a larger sample size and, while highlighting the success of fully automating the CT analysis process, also emphasized the importance of personally viewing the calcification and validating the software’s results before a TAVR procedure.
Outcomes and Implications
When a patient has aortic stenosis, the narrowing or obstruction of the aortic valve, they will often undergo transcatheter aortic valve replacement (TAVR). Computed tomography (CT) analysis is an incredibly important part of the pre-procedure protocol, but human analysis can be time consuming and subject to observer variability and error, so a more accurate solution is needed. In this study, researchers were able to make an AI based software that fully automated the CT analysis process for TAVR preparation that was highly similar to expert analysis, especially when accounting for inter-observer error. This software could reduce the time, expense, and error that comes with current CT analysis in both high and low volume settings. As the study was based on a relatively small sample size, it requires further external validation. If clinically implemented, the authors highly recommend that CTs are still individually observed and validated for the best TAVR results.