EchoNet Cardiac Analysis Tutorial

Left-ventricle segmentation and ejection-fraction prediction from echocardiograms — from setup to results.

Welcome to this tutorial on automated echocardiography analysis. It walks you through two ready-to-run scripts built on top of the EchoNet-Dynamic deep-learning models:

• PredictSegmentation.py — segments the left ventricle (LV) in ultrasound videos and .npy images, measures LV area and detects systolic frames.

• PredictEF.py — predicts the ejection fraction (EF) of the heart directly from an apical-4-chamber ultrasound video.

• CalculateStats.py — (optional) computes the per-channel MEAN/STD used to normalise inputs, so you can retune the models for your own dataset.

Note

Both scripts run on Windows or Linux, with or without an NVIDIA GPU. A GPU is strongly recommended for video processing but not required.

---

📖 Tutorial Contents

• 1. Introduction
  • 1.1. What this tutorial covers
  • 1.2. Background: what the models do
  • 1.3. The end-to-end workflow
  • 1.4. How the scripts are configured
  • 1.5. What you need before starting
• 2. Installation & Setup
  • 2.1. Create a Python environment
  • 2.2. Install PyTorch
  • 2.3. Install the remaining dependencies
  • 2.4. Obtain the pretrained weights
  • 2.5. The project layout
• 3. Data Preparation
  • 3.1. Accepted input formats
  • 3.2. How each script scans the folder
  • 3.3. Image / video assumptions
  • 3.4. Normalisation constants
  • 3.5. Project folder layout
• 4. Computing Normalization Statistics
  • 4.1. Configure the paths
  • 4.2. Two modes: training-only vs full scan
  • 4.3. Run the script
  • 4.4. Use the result
• 5. Left-Ventricle Segmentation
  • 5.1. Configure the paths
  • 5.2. Run the script
  • 5.3. What happens under the hood
  • 5.4. Outputs
• 6. Ejection-Fraction Prediction
  • 6.1. Configure the paths
  • 6.2. Run the script
  • 6.3. What happens under the hood
  • 6.4. Output
• 7. Results & Troubleshooting
  • 7.1. Validating against the EchoNet-Dynamic ground truth
  • 7.2. Common errors
  • 7.3. Performance tips
  • 7.4. Reproducibility checklist
• 8. Downloads
  • 8.1. Scripts
  • 8.2. Pretrained weights
  • 8.3. Sample data (demo)
  • 8.4. Ground-truth labels (for validation)
  • 8.5. After downloading

---

📚 Appendices

• Glossary
• FAQ — Frequently Asked Questions

---

Tutorial Overview

#	Chapter	What you will learn
1	Introduction	The two tools, EchoNet-Dynamic background, the workflow
2	Installation	Conda environment, dependencies, model weights
3	Data Preparation	Accepted input formats (videos and .npy), folder layout
4	Normalization Statistics	Computing MEAN/STD for your own data with CalculateStats.py
5	LV Segmentation	Running PredictSegmentation.py, reading its outputs
6	EF Prediction	Running PredictEF.py, interpreting EF values
7	Results & Troubleshooting	Output files, common errors and how to fix them
8	Downloads	The three scripts and two pretrained weights, ready to download

---

Indices and tables

• Index

• Search Page