CT angiography prior to TAVI has fundamental importance, to proper plan valve dimensions and access sites. Several medical specialists are involved in the procedure planning, namely radiologists, cardiologists and heart surgeons. It is difficult and time consuming to perform manual measurements at every aortic landmark, due to the necessity of manually aligned multiplanar reconstructions. Schmidkonz et al. reported significant interobserver variability of CT for evaluation of aortic dimensions in these patients [10]. In the last years several AI algorithms have been developed to help radiologists to perform different image evaluations; Horehledova et al. demonstrated that semi-automatic analysis with a dedicated software (Syngo.CT Cardiac Planning, Syngo.via VB20A, Siemens) give comparable measurements at CT prior to TAVI with lower reading time [11].
In the current study we assessed the performance of AI-Rad Companion Chest CT (Siemens) for the evaluation of aorta diameters on CT angiography prior to TAVI. Our gold-standards were the diameters manually taken in consensus by two radiologists with expertise in vascular imaging. The AI algorithm demonstrated high accuracy for every aortic landmark, being the values automatically obtained never significantly different from the manual measurements. It is noteworthy that the accuracy, although always high, was not the same for every position: the more the aortic landmark was distal, the more accurate was the automatic analysis. Conversely p-values were lower close to the heart. The reason is the complex anatomy of the aortic root, where the sharp tapering of the sinotubular junction follows the large sinus of Valsalva; in addition, at this level the aorta begins to describe an arch with a radius of varying amplitude. This structural complexity makes difficult for the software to identify the proper plane for measurements. According to our experience, it would be useful a tool that allows the user to take measurements directly on the display of the Vascular module, with the possibility to edit the proposed planes and contours.
As expected, the software loading time was lower compared to the manual measurement time (1 min 47 sec Versus 5 min 41 sec). Before to sign the report it is mandatory to check the automatically obtained values, so the effective time for reporting with the software in the everyday practice would be higher. Anyway as stated by Rueckel et al. AI assisted reporting promises significant time saving compared with fully manual analysis [5].
In order to perform a complete planning of TAVI procedure, additional anatomy details have to be described, as recommended by the expert consensus guidelines [12, 13]. First, it is crucial to report dimensions of the aortic annulus, for accurate prosthesis sizing in order to avoid prosthesis-annulus mismatch (Fig. 4); a small prosthesis could cause dislodgement or paravalvular regurgitation, while a too large could result in annulus rupture [10]. Second, diameters of common and external iliac arteries have a key role, being the iliac access the preferred choice for valve insertion [14]. These information are not yet provided by the Vascular module of Chest CT extension. The radiologist's evaluation of CT images remains fundamental: a comprehensive extra-vascular analysis of chest structures and abdominal organs is mandatory. In addition, motion and misregistration artifacts could compromise AI results. However, a precise and complete software could greatly reduce the time and complexity of reporting in the future.