Introduction: Calcium (Ca) score of the aortic valve has emerged as a tool for assessing aortic valve stenosis severity. The use of computed tomography (CT) is limited due to ionizing radiation and availability. Ca identification based on echo pixels using artificial intelligence (AI) systems has shown promising results in transthoracic echocardiography (TTE). Nevertheless, this technique is highly dependent on the patient’s acoustic window. We propose that transesophageal echocardiography (TEE) obviates the poor acoustic window and could be used for sequential follow-up of patients since no ionizing radiation is used. We also propose that 3D TEE could be more adequate to total valve calcium burden because it allows for identification of calcium pixels in a greater portion of the aortic valve, when compared to 2D techniques. Objective: We aimed to perform an exploratory study to assess the feasibility of quantification of the calcium burden of the aortic valve imaged by 3D TEE and AI. Materials and methods: Prospective analysis. Population: 24 individuals, 14 males, 23 patients with moderate or severe aortic stenosis by TTE, median age 76 years (IQR 12), 1 subject with normal aortic valve for validation of the model. Standard method for Ca scoring: CT scan. Imagiologic studies: TEE exam: 3D volume sets acquired in 3D zoom at the level of the aortic valve, stored in DICOM for post-processing. In the MPR quantification software contiguous 1.5 mm slices were obtained of the aortic valve in diastole, in short axis view. A Computer Vision model was applied to echocardiographic images, via adaptive image segmentation and Deep Learning to identify speckles and artifacts generated by the presence of Ca. The concordance of the Ca speckles of the 3D TEE images and the Agatston score was compared. An AI_Ca_score was obtained by the sum of the pixels of the Ca speckles of the 11 frames of each patients valve. Results: The delay between TEE and CT scans was 65+-47 days. We found a significant positive correlation of the AI_Ca_score with the CT Agatston Ca_score: R= 0,65 (p<0,001, CI 95 % 0,43-0,8). The ROC curve analysis to detect very likely or likely severe calcification showed a very good result with an AUC of 0,86 (C.I: 0,68-1,05) for a cutoff of 68812,5 in the AI Ca Score , with a Sensibility of 90% and Specificity of 75 % and a good result to detect very likely severe calcification with an AUC of 0,75 (C.I: 0,54-0,95) (fig.8B) for a cutoff of 401633 in the AI Ca Score, with a Sensibility of 71% and Specificity of 65%. Conclusions: identification of calcification of the aortic valve by AI from TEE 3D images is feasible and correlates positively with CT scans with a good performance do detect severe calcification. This model should be applied to further ranges of aortic valve calcification to better discriminate severity of the disease.

--