Smart Catheter

Official implementation of "Roughly Collected Dataset for Contact Force Sensing Catheter". All property belongs to Asan Medical Center, Asan Institue of Life Sciences, Biomedical Engineering Research Center for Life Sciences.

• Paper: https://arxiv.org/abs/2102.01932
• Dataset: https://termlab.github.io

Train New Model

If you want to train your own model which is not in our implementations, follow the instructions.

Create package

First, you need to create a new python package. It is done by copying one of the existing package.

cp -r models/fcn models/<your-model>

Then your task package will have the following structure.

models/<your-model>
├── __init__.py
├── arguments.py
├── model.py
└── dataset.py

Implement dataset

In dataset.py, there is Dataset class that is used to fetch crowd-sourcing data. You need to implement __init__(), __len__(), and __getitem__(). While implementing, you should not change the signature of the functions. For each function, there are some requirements you should meet.

• __len__() : return the number of data.
• __getitem__() : return the data which its index is idx, and any shape doesn't matters

If you need additional arguments for dataset, modify add_dataset_args(). Additional information is introduced at argparse.

Implement model

In model.py, there is Model class that is used for predicting contact force from signal data. It is same with other pytorch models, but one restriction exists - the output shape must be (batch_size, 1). Input variable x in forward function is the data you fetched from Dataset instance. As same as Dataset, if you want more arguments, modify add_model_args().

Add to global arguments

After finishing Dataset and Model, please add your task to arguments.py.

# models = ['fcn', 'rnn', 'transformer']
models = ['<your-model>', 'fcn', 'rnn', 'transformer']

Now you can train your own model!

Train

Training is simple. Run python train.py --model <your-model> -h to see the arguments and then add them. Because the arguments for each task is different, you should add --task option to see help. After adding the arguments, just run the command! Logs will be saved as tensorboard log that you can see on tensorboard.

> python train.py --model fcn -h
usage: train.py [-h] [--seed SEED] [--epochs EPOCHS] [--batch_size BATCH_SIZE] [--lr LR] [--log_interval LOG_INTERVAL] [--model {fcn,rnn,transformer}] [--train_data TRAIN_DATA] [--valid_data VALID_DATA] [--device {cpu,cuda}] [--save_dir SAVE_DIR] [--log_dir LOG_DIR] [--input_len INPUT_LEN]
                [--n_channel N_CHANNEL] [--n_hid N_HID] [--n_layer N_LAYER] [--dropout DROPOUT]

optional arguments:
  -h, --help            show this help message and exit

train:
  --seed SEED           Random seed.
  --epochs EPOCHS       Number of epochs for training.
  --batch_size BATCH_SIZE
                        Number of instances in a batch.
  --lr LR               Learning rate.
  --log_interval LOG_INTERVAL
                        Log interval.
  --model {fcn,rnn,transformer}
                        Task name for training.
  --train_data TRAIN_DATA
                        Root directory of train data.
  --valid_data VALID_DATA
                        Root directory of validation data.
  --device {cpu,cuda}   Device going to use for training.
  --save_dir SAVE_DIR   Folder going to save model checkpoints.
  --log_dir LOG_DIR     Folder going to save logs.

dataset:
  --input_len INPUT_LEN
                        Lenght of input signal.
  --n_channel N_CHANNEL
                        Number of channels in input signal.

model:
  --n_hid N_HID         Number of units in a FC layer.
  --n_layer N_LAYER     Number of FC layers.
  --dropout DROPOUT     Dropout rate.

Test

With our checkpoint file, you can measure performance on test set. Run test.py with your checkpoint file, than you can see the L1 loss and 95% reliability.

> python test.py -h
usage: test.py [-h] [--batch_size BATCH_SIZE] [--test_data TEST_DATA] [--device {cpu,cuda}] [--ckpt_dir CKPT_DIR]

optional arguments:
  -h, --help            show this help message and exit

test:
  --batch_size BATCH_SIZE
                        Number of instances in a batch.
  --test_data TEST_DATA
                        Root directory of test data.
  --device {cpu,cuda}   Device going to use for training.
  --ckpt_dir CKPT_DIR   Directory which contains the checkpoint and args.json.

Experiments

With several combinations of # of hidden, # of layers (and # of heads for transformer), the following models gain the best performance for test data. We used preprocessed data, not the original data. You can see preprocess.py if you're interested in our preprocessing.

• FCN: 2 layers, 64 hidden dimension
• RNN: GRU cell, 4 layers, 64 hidden dimension
• Transformer: 12 heads, 4 layers, 512 hidden dimension

Model	FCN	RNN	Transformer
Performance	3.03g	2.46g	3.01g

Citations

TODO