Truace
Subscribe
55/100

Moderate evidence

Cite
More
Export CSVExport JSONEmailLinkedInX

Heart ultrasound AI graded blood-stasis 'smoke' with 93.7% accuracy in a study of 1,106 videos

Study title: SECNet: Differentiable Mask-Driven Spatiotemporal Learning for Video-Based Spontaneous Echo Contrast Severity Detection

Evidence type
Benchmark or lab evaluation
Sample
n = 1,106
Sources
1 sourcePeer-reviewed
Published
2 min read
Truvace record
TRV-2026-2467cite or check it

Study title: SECNet: Differentiable Mask-Driven Spatiotemporal Learning for Video-Based Spontaneous Echo Contrast Severity Detection

Researchers built SECNet, a deep learning system that detects and grades spontaneous echo contrast (SEC) in transesophageal echocardiography videos. Trained on 1,106 annotated videos, it reached 93.7% accuracy and a 90.0% F1-score, beating the compared methods by 3.1 to 23.3 percentage points in accuracy.

Read the study abstract

Spontaneous Echo Contrast (SEC), characterized by smoke-like dynamic echogenicity in trans esophageal echocardiography (TEE), serves as a critical imaging biomarker for assessing thromboembolic risk. However, current diagnostic approaches depend on manual interpretation or regression-based algorithms, which are labor-intensive, prone to inconsistency, and limited in generalizability. To address these challenges, we propose SECNet, an end-to-end deep learning framework for auto mated SEC detection in TEE videos.…

The gain

In a study of 1,106 annotated transesophageal echocardiography videos, the SECNet model graded spontaneous echo contrast severity with 93.7% accuracy, 3.1 to 23.3 percentage points above the compared methods.

In brief

SEC appears as smoke-like moving echoes on heart ultrasound and is used as an imaging marker of clotting risk. Current assessment relies on manual reading or regression-based algorithms, which the authors describe as labor-intensive, inconsistent and limited in how well they generalize. The authors say the tool supports standardized, reproducible assessment to assist clotting-risk evaluation in clinical workflows.

Main points

  1. SECNet was trained on a clinical dataset of 1,106 annotated transesophageal echocardiography videos.
  2. It reached 93.7% accuracy and a 90.0% F1-score for SEC severity detection.
  3. It beat the compared methods by 3.1 to 23.3 percentage points in accuracy.
  4. One module segments the heart regions where SEC tends to appear and suppresses surrounding noise.
  5. A second module captures how SEC moves over time and where it is distributed in the image.
  6. A differentiable mask links the two modules so localization and classification are optimized together.
  7. A further experiment on the EchoNet Dynamic dataset tested the architecture on a related echocardiographic video task.

The full story

SECNet has two parts: a region-of-interest segmentation module that localizes cardiac structures prone to SEC while filtering out surrounding noise, and a spatiotemporal feature extraction module that captures both the motion and the spatial spread of the smoke-like pattern.

The two modules are connected by a differentiable mask-guided bridge, which lets the localization and the severity classification be trained jointly rather than as separate steps.

The authors also ran an auxiliary experiment on the public EchoNet Dynamic dataset, which they say shows the architecture can apply to a related echocardiographic video-analysis task. They say the approach could support standardized, reproducible SEC assessment in clinical workflows.

Talk it overFour questions for a class, a team or the dinner table

  1. Who gets this benefit, and who was left out of the study or story?

    The gain In a study of 1,106 annotated transesophageal echocardiography videos, the SECNet model graded spontaneous echo contrast severity with 93.7% accuracy, 3.1 to 23.3 percentage points above the compared methods.

  2. How do we know? What can this kind of evidence show, and what can’t it?

    The evidence This rests on a benchmark or lab test with a sample of 1,106, from IEEE Journal of Biomedical and Health Informatics.

  3. What does it leave open? What else would you want to know before acting on it?

  4. Have you seen AI used this way where you study, work or live? Did it go the same way?

The questions are the same on every record. The details under them are taken from this record as written.

Sources

1 on file
  1. IEEE Journal of Biomedical and Health Informaticsdoi:10.1109/jbhi.2026.3751319
    Peer-reviewed2026-11-06

The debate

More from Health