Artificial Intelligence-Based Echocardiography in Pulmonary Arterial Hypertension

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Echocardiography is crucial for pulmonary hypertension (PH) assessment.
• Manual interpretation is time-consuming and error-prone.
• Automated solutions are needed to improve efficiency and accuracy.

Purpose of the Study:

• To evaluate the reliability of a fully automated deep learning (DL) workflow in echocardiography for PH assessment.
• To determine the bias and precision of DL interpretations compared to core laboratory readers.
• To assess DL's ability to discriminate milder forms of PH.

Main Methods:

• Two study parts: bias/precision assessment and discrimination of milder PH.
• Used Us2.ai software version 1.4.5 with core laboratory (CL) readers as reference.
• Included case-control and referral cohorts with parameters like peak tricuspid regurgitation velocity (TRV) and right ventricular (RV) function.

Main Results:

• No significant bias for peak TRV, RA area, and TAPSE; RVFAC showed significant bias (11.46%).
• Precision for most measurements was below 15%.
• Area under the curve (AUC) for PH detection was 0.98 for DL vs. 0.80 for CL in one cohort, and 0.75 for DL vs. 0.79 for CL in the referral cohort.

Conclusions:

• The automated DL workflow is promising for PH assessment in echocardiography.
• It has the potential to enhance efficiency in clinical practice.
• Further validation, particularly for RVFAC, may be necessary.