embryo-analysys

Environment

python : 3.6.9

PyTorch : 1.4.0

PyTorch vision : 0.5.0

Installation

Installing PyText and making virtualenv

how to install Pytext:

https://github.com/facebookresearch/pytext

after making virtualenv, install requirement.txt

pip install -r requirements.txt

Execution

frame-wise

1. make csv

python make_csv.py --time 25 --data_dir "/home/masashi_nagaya/M2/dataset_9_08/all/"

2. train

pn classification

python abn_pn.py --resume ./checkpoint/imagenet/resnet50/checkpoint.pth.tar --val_num 1 --manualSeed 1 –a resnet50 

nnpu classification

python abn_pu.py --resume ./checkpoint/imagenet/resnet50/checkpoint.pth.tar --val_num 1 --manualSeed 1 –a resnet50

auc-pr optimization

python abn_auc-pr.py --resume ./checkpoint/imagenet/resnet50/checkpoint.pth.tar --val_num 1 --manualSeed 1 –a resnet50

auc-pr-nnpu optimization

python abn_auc-pr-pu.py --resume ./checkpoint/imagenet/resnet50/checkpoint.pth.tar --val_num 1 --manualSeed 1 –a resnet50

when mutual information maximization, change codes as follows:

• PN：abn_pn.py→abn_pn_iic.py
• nnPU：abn_pu.py→abn_pu_iic.py
• AUC-PR：abn_auc-pr.py→abn_auc_pr_iic.py
• AUC-PR-nnPU：auc-pr-pu.py→ auc-pr-pu_iic.py

3. test

all-frame evaluation

python test.py --resume ./checkpoint/pn_mse1.0_seed --val_num 1　--seed_number 5 --mode pn --eval all

final-frame evaluation

python test.py --resume ./checkpoint/pn_mse1.0_seed --val_num 1　--seed_number 5 --mode pn --eval last

weighted average evaluation

python test_weighted.py --resume ./checkpoint/pn_mse1.0_seed --val_num 1 --seed_number 5 --mode pn

attention map output

python attention.py --resume ./checkpoint/pn_mse1.0_seed1.pth.tar --val_num 1 --test_mode pn

Recurrent

1. make csv

python make_csv.py

2. feature extraction

python extract_aug.py --val_num 1 --resume ./checkpoint/pn1_mse1.0_seed1.pth.tar 

3. train

final-state feature

python gru_final.py --val_num 1 --manualSeed 1

temporal-pool feature

python gru_temporal.py --val_num 1 --manualSeed 1

4. test

python gru_eval.py --val_num 1 --resume ./checkpoint/gru1_seed --arch final 

Citation

Please cite this paper in your publications if this dataset helps your research.

@article{
  title={Embryo grading from unreliable labels by positive-unlabeled classification with ranking},
  author={Masashi Nagaya and Norimichi Ukita},
  journal={{IEEE} Transactions on Medical Imaging},
  volume    = {41},
  number    = {2},
  pages     = {320--331},
  year={2022}
}