Research Project
Phase 1: Data Collection

AI Phenotype Identification

Building an AI model to automatically identify hemodynamic phenotypes from EASy POCUS images

About This Project

Training AI with Expert-Labeled Data

We are building a machine learning model that can automatically identify hemodynamic phenotypes from EASy POCUS images. This requires high-quality, expert-labeled training data.

Users upload video clips, then POCUS experts review and label each submission with the correct phenotype and quality score. These expert-labeled video clips continuously train our AI model to recognize patterns associated with each of the 10 EASy MAP phenotypes.

1

User Upload

Users submit video clips

Ongoing
2

Expert Labeling

Experts label phenotype & quality

Ongoing
3

Continuous Training

AI learns from new labeled data

4

AI Deployment

Automated phenotype identification

Progress

Dataset Building Progress

Our goal is 5,000+ labeled video clips across all 10 phenotypes

847
Video Clips
Goal: 5,000
12
Contributing Experts
4
Institutions

Video Clips by Phenotype

1Hypovolemic Shock
162 / 250
2Distributive Shock
201 / 250
3Diastolic Dysfunction
62 / 250
4Cardiogenic Shock (LV)
138 / 250
5Biventricular Failure
76 / 250
6Acute RV Failure
99 / 250
7Acute on Chronic RV Failure
38 / 250
8Pericardial Tamponade
24 / 250
9Catastrophic Valve Disease
31 / 250
10Tension Pneumothorax / Auto-PEEP
16 / 250
We need more images for phenotypes 7, 8, 9, and 10. If you have cases with these phenotypes, please contribute!
Expert Contribution

Upload & Label Video Clips

Upload your EASy POCUS video clips and provide the expert phenotype label

1. Users Upload

Submit video clips

2. Experts Label

Identify phenotype & quality

3. AI Learns

Continuously trains on data

Subcostal 4-ChamberUpload video clip
IVC ViewUpload video clip
Right Upper LungUpload video clip
Left Upper LungUpload video clip
Right PleuralUpload video clip
Left PleuralUpload video clip

Select Phenotype (Expert Label)

Choose the hemodynamic phenotype based on your expert interpretation

Cluster 1: Small/Normal Cavity, Normal/Increased Contractility
1Hypovolemic Shock
2Distributive Shock
3Diastolic Dysfunction
Cluster 2: Enlarged LV +/- RV, Decreased Contractility
4Cardiogenic Shock (LV)
5Biventricular Failure
Cluster 3: Isolated Enlarged RV
6Acute RV Failure
7Acute on Chronic RV Failure
Cluster 4: Obstructive Causes
8Pericardial Tamponade
9Catastrophic Valve Disease
10Tension Pneumothorax / Auto-PEEP
Coming Soon

AI-Powered Phenotype Identification

Once we have sufficient labeled data, we will train and deploy an AI model that can:

  • Automatically analyze uploaded EASy POCUS video clips
  • Identify hemodynamic phenotype with confidence score
  • Provide treatment suggestions based on phenotype
  • Integrate with EASy Sepsis app for real-time guidance
Estimated Launch: When dataset reaches 5,000+ labeled video clips
AI Analysis Result
4Cardiogenic Shock (LV)
Confidence: 94%
Suggested: Inotropic support, limit fluids

Become a Contributor

We are looking for POCUS experts to help build our training dataset. If you have experience with hemodynamic assessment and access to de-identified video clips, join our research team.