Prediction of mortality by echocardiographic parameters: human vs. fully-automated AI-based analysis

Background:

The use of artificial intelligence (AI) for echocardiography analysis has been rapidly advancing and integrating into clinical practice. AI applications have demonstrated good accuracy in validation models compared to human experts. It is unclear how fully automated echocardiographic measures correlate with clinical outcomes, as compared to human readers.

Purpose

We aim to compare the performance of AI-based echocardiographic measurements in prediction of mortality as compared with human expert analysis.

Methods

We retrospectively analyzed 889 consecutive hospitalized patients who underwent a clinically indicated transthoracic echocardiographic study. Each study was performed by experienced cardiac sonographers and reviewed by expert echocardiologists. All studies were also analyzed by a commercial, fully automated AI software (Us2.AI). For both human and AI measurements, we constructed multivariable models to predict 1-year all-cause mortality. Variables included in the models were selected based on their significance in univariate analysis. Cut-offs values were identified using a decision tree and the Chi-squared Automatic Interaction Detector (CHAID) algorithm. We used receiver operating characteristic (ROC) curves to determine which model provided stronger predictive power for mortality.

Results

Of the 889 patients, 731 (82%) patients (mean age 68±16, 46% Female) had sufficient echocardiographic data to be included in the analysis. One-year mortality was 13% (95 patients). Table 1 shows univariate analysis of human and AI measurements for 1-year all-cause mortality. 15 parameters automatically obtained by the AI system were found to be correlated with mortality, while 9 parameters from the human-based analysis were significant. In the multivariable regression model for 1-year mortality (Figure 1), the AI model showed superior predictive value (ROC area under the curve =0.758 vs. 0.709, P=0.041).

Conclusion

In this cohort of hospitalized patients, fully automated AI-based echocardiographic analysis was superior to human expert analysis in prediction of 1-year mortality.