AS_PLAX

This repo contains code and model for paper Identifying Aortic Stenosis with a single PLAX video

Trained Model Weights

The trained model check points are shared in the dropbox folder: https://www.dropbox.com/sh/vg0avrsrkkw4ipl/AADkZAZxW73VdxJ6gZG_OXQ0a?dl=0

MA: ./ma_checkpoint

MP: ./mp_checkpoint

MS: ./ms_checkpoint

Dependencies

Dependencies are defined in as_plax.yml, to create a conda environment:

conda env create -f echo_models.yml

Preprocessing of Echo Data

The preprocessing includes extracting pixel files from the raw Dicom format, de-identification, removal of ECG and other meta data and resizing/sampling the frames of echo videos. Source codes are under ./utils.

Creating a Dataset

To load a new dataset for training and inference, create a dataset file under ./datasets like dataset_example.py.

Training a Model

To train a model, create a new config file under ./configs like train_example.json.

In this config file, define training configerations like path to data, label, hyperparamters and logging options, etc.

Command to launch training: python train.py --json_file path/to/json

Testing a Model

To test a model, create a new config file under ./configs like test_example.json.

In this config file, define testing configerations like path to data, ground truth label, model checkpoint and logging options, etc.

Command to launch testing: python test.py --json_file path/to/json