Automated mitral inflow Doppler peak velocity measurement using deep learning.

Area of Science:

• Cardiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Biomedical Signal Processing

Background:

• Doppler echocardiography is crucial for non-invasive assessment of heart valve function, particularly the mitral valve.
• Manual analysis of Doppler traces introduces significant inter- and intra-observer variability, necessitating automated solutions.
• Accurate peak velocity measurements are vital for diagnosing and managing valvular heart disease.

Purpose of the Study:

• To develop and validate a deep learning model for automated detection of peak velocity measurements from mitral inflow Doppler images.
• To create an automated system independent of electrocardiogram (ECG) signals for enhanced robustness.
• To establish a benchmark dataset and open-source model to facilitate further research and clinical translation.

Main Methods:

• Development of a deep learning model utilizing heatmap regression networks for automated peak velocity detection.
• Creation of a comprehensive dataset of mitral inflow Doppler images annotated by multiple expert cardiologists.
• Evaluation of the model's performance against expert consensus, assessing its accuracy and variability.

Main Results:

• The deep learning model achieved a 96% detection accuracy for peak velocity measurements.
• The model's discrepancy with expert consensus was within the established range of inter- and intra-observer variability.
• The developed dataset and models are made open-source to promote wider adoption and further development.

Conclusions:

• The proposed deep learning model offers a reliable and accurate automated solution for peak velocity measurement in mitral inflow Doppler echocardiography.
• This automated approach effectively minimizes the variability associated with manual assessments, enhancing diagnostic consistency.
• The open-source availability of the dataset and model encourages further innovation and clinical integration of AI in cardiovascular imaging.