EchoNet Function Evaluation:
Quantifying Variation in Assessing Cardiac Function

An automated end-to-end pipeline for

1. assessing cardiac function
2. quantifying error in left ventricular ejection fraction calculation with slight clinical perturbations

Dataset

We simulated errors and ran calculations on 3,906 echocardiogram videos from EchoNet-Dynamic (https://echonet.github.io/dynamic/)

Examples

We simulated variation in 3,906 apical-4-chamber echocardiogram videos by automatically generating the equivalent of 976,500 left ventricular tracings with varying degrees of individual tracing errors introduced at end systole. The clinical variation showcased are over/under tracing of the left ventricle, malrotations of main longitudinal axis, foreshortening at the level of the apex and mitral valve.

Normal Tracing	Over/Under Tracing	Angle Rotations	Main Axis Foreshortening

Installation

First, clone this repository and enter the directory by running:

git clone https://github.com/douyang/echonet-function-evaluation
cd echonet-function-evaluation

EchoNet-Function-Evaluation is implemented for Python 3, and depends on the following packages:

• NumPy
• Matplotlib
• Pandas
• OpenCV
• Pillow
• SciPy
• tqdm

The dependencies can be installed by navigating to the cloned directory and running:

pip install --user .

Usage

Setting Path to Data

By default, the pipeline assumes that a copy of the data is saved in a folder named EchoData/ in this directory. This path can be changed by setting an absolute path in the file named data.cfg (an example is given in the file already).

Running Code

The pipeline has two main tasks: assessing cardiac function and quantifying variation on left ventricle ejection fraction calculations. Each of these tasks can be run with the scripts below.

Predicting Cardiac Timings

cmd="import echofunctions; echofunctions.timing_predictions.gatherTimingPredictons(inputFolder=\"Videos-Segmented\",
                                                            outputFolder=\"all_frames\", 
                                                            outputCSVName=\"Frame Timing Predictions.csv\",
                                                            createFrames=False)"
python3 -c "${cmd}"

We offer a solution to finding the appropriate end-systolic and end-diastolic timings from a given segmented echocardiogram video. The above command will output individual frames of each video in the dataset to a folder. It will also create a CSV with the cardiac timing predictions of the pipeline and of human assessment for reference.

Outputting Frames

cmd="import echofunctions; echofunctions.create_frames.exportFrames(segmentedVideosFolder=\"red_frames\")"
python3 -c "${cmd}"

This will output the appropriate frames of each video from a given CSV that contains the end-systolic and end-diastolic timings.

Creating Masks

cmd="import echofunctions; echofunctions.mask.generateMasks(segmentedFramesFolder=\"red_frames\",
                                                      outputMasksPath=\"algorithm_masks\")"
python3 -c "${cmd}"

This will create a visual representation of the volumetric calculation for each segmented image in a given folder. The masks will be outputted to any preferred folder.

Running Ablations

cmd="import echofunctions; echofunctions.ablations.exportCSV(inputFolderName=\"red_frames\", 
                                                            fileName=\"ablations.csv\",
                                                            task=\"Angle Shift\")"
python3 -c "${cmd}"

There are various ablations that can be run to systematically mimic and quanity left ventricular ejection fraction in clinical procedures. This script will run ablations on a set of images in a folder and export a CSV with the volumetric data with the ablations.

Generating Plots

The final analyses can be run with the various scripts provided in the folder named scripts. These scripts will utilize the results from ablations.csv and create visuals. Different statistics can also be calculated from the standard volume data in Base Volume.csv to assess the variation generated by the pertubations.

Mask Sweeps

The full visuals of the ablations can be outputted by sweeping through segmented images in a given folder. The mask_sweeps.py script will create generate frames with example images including each data input ablation. In particular, the different parameters include the test videos (number of videos to assess), the method of volume calculation (Method of Disks, Bullet Method, Prolate Ellipsoid), and analysis type (over and under tracing, rotations in main axis longitudinal axis, and mitral valve annulus level error.