Code for the MLHC 2021 paper: A New Semi-supervised Learning Benchmark for Classifying View and Diagnosing Aortic Stenosis from Echocardiograms
Please visit our website https://tmed.cs.tufts.edu/index.html and download the data
Follow the instructions here: https://conda.io/projects/conda/en/latest/user-guide/install/index.html
conda env create -f environment.yml
Folder Data_Processing provides script to process the raw png files into tfrecords, which can then be loaded to train the model. For example, if you want to create the tfrecord according to our suggested split1 on TMED-18-18
bash process_TMED_18_18_fold0.sh
The commands for reproducing major tables in the paper are provided in runs
For example, if you want to run a fully supervised baseline on our suggested split1 on the TMED-18-18
bash launch_fs_fold0.sh run_here
We have provided the saved image level predictions in folder predictions, so that the patient level prediction code can run smoothly to reproduce our results in the paper.
For example, if you want to run the patient level prediction for MixMatch on our suggested split1 on TMED-156-52, go to runs/table7
bash launch_MixMatch_fold0.sh run_here
In practice, the user can save their own image level predictions to the folder and run the script instead.
The results including the training curves, validation and test balanced accuracy with and without ensemble etc, will be automatically saved under the results_analysis folder (created during training) under your specified train_dir in the bash script.
While the focus of our paper is reproducibility, ultimately exact comparison to the results in our paper will be conflated by subtle differences such as the version of TensorFlow used, random seeds, etc.
@inproceedings{huangSemisupervisedEchocardiogramBenchmark2021, title = {A New Semi-supervised Learning Benchmark for Classifying View and Diagnosing Aortic Stenosis from Echocardiograms}, booktitle = {Machine Learning for Healthcare}, author = {Huang, Zhe and Long, Gary and Wessler, Benjamin and Hughes, Michael C.}, year = {2021}, }