Cyclical Self-Supervision for Semi-Supervised Ejection Fraction Prediction from Echocardiogram Videos

This is the implementation of CSS for Semi-Supervised Ejection Fraction Prediction for the paper ["Cyclical Self-Supervision for Semi-Supervised Ejection Fraction Prediction from Echocardiogram Videos"]().

Data

Researchers can request the EchoNet-Dynamic dataset at https://echonet.github.io/dynamic/ and set the directory path in the configuration file, echonet.cfg.

Environment

It is recommended to use PyTorch conda environments for running the program. A requirements file has been included.

Training and testing

The code must first be installed by running

pip3 install --user .

under the repository directory CSS-SemiVideo. Training consists of three components:

1) To train the CSS semi-supervised segmentation model, run:

echonet seg_cycle --batch_size=20 --output=output/css_seg --loss_cyc_w=0.01 --num_epochs=25 --rd_label=920 --rd_unlabel=6440 --run_test --reduced_set 

The LV segmentation prediction masks of all frames must be inferred for the second stage. To do so, run:

echonet seg_cycle --batch_size=20 --output=output/css_seg --loss_cyc_w=0.01 --num_epochs=25 --rd_label=920 --rd_unlabel=6440 --skip_test --reduced_set --run_inference=train

echonet seg_cycle --batch_size=20 --output=output/css_seg --loss_cyc_w=0.01 --num_epochs=25 --rd_label=920 --rd_unlabel=6440 --skip_test --reduced_set --run_inference=val

echonet seg_cycle --batch_size=20 --output=output/css_seg --loss_cyc_w=0.01 --num_epochs=25 --rd_label=920 --rd_unlabel=6440 --skip_test --reduced_set --run_inference=test

The segmentation prediction outputs will be located under the output folder output/css_seg. To reduce installation time for EchoNet-Dynamic, these are moved to a separatate directory parallel to CSS-SemiVideo, i.e. CSS-SemiVideo/../infer_buffers/css_seg. Segmentation masks are also sourced from this location for Step 2 of the framework.

To do this, run:

mkdir ../infer_buffers/css_seg
mv output/css_seg/*_infer_cmpct ../infer_buffers/css_seg/

2) To train the multi-modal LVEF prediction model, run:

echonet video_segin --frames=32 --model_name=r2plus1d_18 --period=2 --batch_size=20 --output=output/teacher_model --num_epochs=25 --rd_label=920 --rd_unlabel=6440 --run_test --segsource=css_seg

3) To train teacher-student distillation, run:

echonet vidsegin_teachstd_kd --frames=32 --model_name=r2plus1d_18 --period=2 --batch_size=20 --output=output/end2end_model --num_epochs=25 --rd_label=920 --rd_unlabel=6440 --run_test --reduced_set --max_block=20 --segsource=css_seg --w_unlb=5 --batch_size_unlb=10 --weights_0=output/teacher_model/best.pt 

Pretrained models

Trained checkpoints and models can be downloaded from:

1) CSS for semi-supervised segmentation: https://hkustconnect-my.sharepoint.com/:f:/g/personal/wdaiaj_connect_ust_hk/EqiP-N0MDRZGlwqr5PeZUrYBtLki8QWtBlMqRK1FNkjbcw?e=DIpkIm

2) Multi-modal LVEF regression: https://hkustconnect-my.sharepoint.com/:f:/g/personal/wdaiaj_connect_ust_hk/ErxaHepi4ndAnMcvSOwTH5wBDI6rHypqdcBiXF8B0XYvmg?e=Rud7Pf

3) Teacher-student distillation: https://hkustconnect-my.sharepoint.com/:f:/g/personal/wdaiaj_connect_ust_hk/Ev7mQ1ReI05LtiDIqQu1IpYBC6xN4R47PsYnhDUQr4n3fw?e=US4caq

To run with the pretrained model weights, replace the .pts files in the target output directory with the downloaded files.

Notes

• Contact: DAI Weihang (wdai03@gmail.com)
  
  

Citation

If this code is useful for your research, please consider citing:

(to be released)