Multiview Echocardiography Videos for Ventricular Abnormality Detection

A collection of 30 echocardiography studies, each with three standard non-Doppler video views. It provides study-level labels for ventricular abnormality and videos preprocessed to 224x224 pixels and 64 frames. The data is intended to illustrate a multiview format and support code reproducibility.

Use Cases

• Train a multiview deep learning model on the three standard echocardiography views to detect study-level ventricular abnormality labels.
• Reproduce model-loading and training code using the provided preprocessed video clips and associated CSV file paths.
• Analyze the relationship between video features from apical four-chamber, two-chamber, and parasternal long-axis views and the ventricular abnormality label.

Strengths

• Includes 30 distinct echocardiography studies, each with three standardized video views.
• Videos are consistently preprocessed: cropped, text removed, resized to 224x224 pixels, and trimmed to 64 frames.
• Data is fully de-identified with no protected health information, facilitating open research.

Limitations

• Small scale with only 30 studies, making it unsuitable for training robust clinical prediction models.
• Limited to study-level labels for ventricular abnormality, lacking frame-level or pixel-level annotations.
• Contains only three specific non-Doppler echocardiography views, not a full clinical echo study.

Provenance

Source

ICPSR Harvested Dataverse

Collection Method

Echocardiography studies preprocessed as described in the associated research manuscript.

Time Range

null

Freshness

null

Geography

null