Deep learning techniques used to segment lungs and metastasis on mice MRI images.
We recommend you to use conda to create your virtual env and manage dependencies.
To install required libraries:
conda env create -f environment.yml
If you do not want to use conda, with python 3.10 and pip:
pip install -f requirements.txt
To measure the performance of each network, we rely on several metrics:
- IoU (Jaccard index)
- AUC (AUROC).
python -m src.train
python -m src.predict
--epochs EPOCHS, -e EPOCHS
number of epochs of training
--batch_size BATCH_SIZE, -bs BATCH_SIZE
size of the batches
--size SIZE, -s SIZE Size of the image, one number
--drop DROP, -d DROP Dropout rate
--filters FILTERS, -f FILTERS
Number of filters in first conv block
-w1 W1 Weight for bg
-w2 W2 Weight for lungs
-w3 W3 Weight for metastases
-lr LR Learning rate.
-nworkers NWORKERS Number of workers in dataloader.
-classes CLASSES Number of classes to predict.
--save If flag, save predictions.
--postprocess If flag, apply postprocess on predictions.
--restart RESTART Restart for cosine annealing. Default 5. Set this to 1 to disable.
--restart_mult RESTART_MULT
A factor increases Ti after a restart. Default 1.
--model {unet,unet_res,unet3p,res_unet3p,urcnn,att_unet,att_unet3p,unet++}
Model name. unet or unet_res
--model_path MODEL_PATH
Weights file name.
--metric {iou,f1} Metric for stats. iou or f1
--loss {ce,focal,tanimoto,lovasz,fusion}
Loss function.
--conv {conv,convsep}
Conv layer to use (conv or convsep).
--contrast If flag, enhance contrast on image.
--img_path IMG_PATH Path to the tiff mouse stack.
--label_path LABEL_PATH
Path to the labels folder, if exist. If not, no stats will be processed.
To generate documentation files and read them :
cd docs
make html
open _build/html/index.html
To test our network rapidly on your images, we advise you to try our plugin for napari, napari-deepmeta. This plugin allows you to try our 2 networks on your images without writing any line of code.