Detect Mitoses cell structures

Machine Learning Model to classify detect Mitoses cell structures

Github: https://github.com/aaljuhani/ML_project

Dataset

URL for TUPAC challenge: http://tupac.tue-image.nl/node/3Direct
URL for the Mitosis Dataset: https://drive.google.com/drive/u/0/folders/0B--ztKW0d17XbXNPQVY5VWxiZkU#list

Installation Instructions

Prerequierment:

Numpy: conda install -c conda-forge numpy npm: conda install -c conda-forge nodejs opencv: pip install opencv=3.4.2 pip install opencv-contrib=3.4.2

create conda env

conda env create -n histosearch

then activate the conda virtual environment

source activate histosearch

To Run the Model

python recognize.py --data <data_DIR> --gt <groundtruth_DIR>

example: python recognize.py --data data/tiles --gt data/groundtruth

To Run the CNN Model on a single node

Tensorflow-GPU and Keras should be installed.

python resnet50.py --data <data_DIR> --gt <groundtruth_DIR> -lbn_points 8 -lbn_r 1

example: python resnet50.py -model nn -d mitoses_image_data/mitoses_image_data/mitoses_image_dataset -gt mitoses_image_data/TUPAC_groundtruth/mitoses_ground_truth/mitoses_ground_truth -lbn_points 8 -lbn_r 1

To run the CNN Model on multiple nodes

Horovod, openMPI, NCCL2 should be installed.

mpirun -np <# of nodes> -hostfile -tag-output -bind-to none -map-by node -x NCCL_DEBUG=INFO -x LD_LIBRARY_PATH -x PATH -mca pml ob1 -mca btl ^openib python resnet50_distributed.py -model nn --data <data_DIR> --gt <groundtruth_DIR> -lbn_points 8 -lbn_r 1

example: mpirun -np 4 -hostfile $PBS_NODEFILE -tag-output -bind-to none -map-by node -x NCCL_DEBUG=INFO -x LD_LIBRARY_PATH -x PATH -mca pml ob1 -mca btl ^openib python resnet50_distributed.py -model nn -d mitoses_image_data/mitoses_image_data/mitoses_image_dataset -gt mitoses_image_data/TUPAC_groundtruth/mitoses_ground_truth/mitoses_ground_truth -lbn_points 8 -lbn_r 1

aaljuhani / ML_project