Automated multi-beat tissue Doppler echocardiography analysis using deep neural networks.

Area of Science:

• Cardiovascular Imaging
• Echocardiography
• Artificial Intelligence in Medicine

Background:

• Tissue Doppler imaging (TDI) is crucial for assessing myocardial blood velocity non-invasively.
• Current manual analysis of TDI is time-consuming and prone to variability.
• Automated identification of heartbeats and peak velocities in TDI is needed to improve efficiency and accuracy.

Purpose of the Study:

• To develop and validate a deep neural network system for automated beat isolation and peak velocity measurement in TDI.
• To compare the performance of the automated system against expert cardiologists and inter-observer variability.
• To assess the efficiency gains of the automated approach compared to manual analysis.

Main Methods:

• A dataset of expert-annotated TDI images was used to train deep neural networks.
• The models were trained for automated prediction of peak Doppler velocities across multiple heartbeats.
• Performance was evaluated using Bland-Altman analysis and compared to expert measurements and inter-observer variability.

Main Results:

• Automated measurements demonstrated good agreement with expert consensus (SD 0.40 cm/s), outperforming inter-observer variability (SD 0.65 cm/s).
• The system's performance was comparable to individual expert operators (SD 0.40-0.75 cm/s).
• The automated approach analyzed over 26 times more heartbeats than manual methods, significantly reducing processing time.

Conclusions:

• The proposed automated system accurately and reliably measures peak Doppler velocities from TDI images.
• The AI's performance is indistinguishable from human experts, offering substantial improvements in processing speed.
• Publicly released datasets and models will facilitate future research and benchmarking in automated echocardiography analysis.