Background

Severe tricuspid regurgitation is associated with poor prognosis. Echocardiography is the method of choice to evaluate valve regurgitation as well as left and right ventricular function. However, estimating cardiac volumes with echocardiography is challenging. Therefore, cardiac MRI (cMRI) has become the Goldstandard. Unfortunately, in many patients cMRI is not feasible. Cardiac CT (cCT) with its high temporal and spatial resolution might be another method of choice.

Methods:

Cardiac volumes of 27 Patients suffering from severe tricuspid regurgitation were analyzed using cardiac MRI and cardiac CT-Scans. MRI-Scans were analyzed from radiologists and were used as Goldstandard. CT-Scans were analyzed by heart.ai, an artificial intelligence software provided by Laralab, Munich Germany. Mean, standard deviation, mean difference, standard deviation of difference (SDD), limits of agreement (LoA) and intraclass correlation coefficients (ICCs) were calculated as well as Bland-Altman plots were constructed using SPSS Statistics 27 from IBM.

Results:

Left ventricular ejection fraction in 27 Patients (63% female, mean age 78 years ) was 57.1% in cMRI and in 57.1% cCT; SDD with LoA 7.8 (15.3 -15.1); ICC 0.884. Right ventricular ejection fraction was 50.5 ±9.8% in cMRI and 51.1 ±7,7% in cCT, SDD 9.15 (17.3; -18.6); ICC 0.637.

Right ventricular stroke volume was 105±29 ml in MRI and 126±27 ml in cCT; SDD 31.18 (-3.7; - 125.9); ICC 0.039. Left ventricular stroke volume was 61.4 ±17 ml average in MRI and 62.9 ±27.5 ml in CT; SDD 12.33 (22.64; -25.69); ICC 0.831. Assuming that no other relevant valve regurgitation was present and cardiac output was steady, regurgitation volume was calculated as difference between left and right ventricular stroke volume. In cMRI calculated regurgitation was 43.7 ±28.2 ml whereas in cCT it was 63.28 ± 29.5 ml; SDD 32 (43.11; - 82.33); ICC 0.483.

Discussion and Conclusion:

Measurements of left and right ejection fraction as well as left ventricular stroke volume using artificial intelligence in CT-Scans provided highly accurate estimations compared with MRI data, whereas right stroke volume und regurgitation volume correlated less. This might be attributed to a certain lay between the CT and MRI scan in a few patients with a slightly different hydratation status. Furthermore, AI-based cCT eavluation might be more influenced by atrial fibrillation as compared to cMRI.

In summary, artificial intelligence software seems to be a fast and accurate way to evaluate cardiac volumes from a routinely obtained full-cycle CT scan. Further studies are warranted to confirm these results.