ArtechStark / OncoNet_Public

Developing Deep Learning Models for Mammography

Home Page:http://learningtocure.csail.mit.edu

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

OncoNet: Developing Deep Learning Models for Mammography

Introduction

This repository was used to develop the models described in:

As the repository is updated with new papers, we will create a new branch for each.

Usage

This code base is provided for clarify implementation details. It is not runnable in a stand-alone fashion since it assumes access to images/metafiles to initialize the Dataset object. If you want to try the models described in the papers above, please checkout OncoServe, our model deployment codebase. It supports running all of our published models with a webserver HTTP interface.

Command to train the Density Model:

CUDA_VISIBLE_DEVICES=0 python -u scripts/main.py  --batch_size 32 --cuda --dataset mgh_mammo_full_density --dropout 0.4 --epochs 100 --img_dir /scratch1/mammosprint --img_mean 7662.576866173061 --img_std 12594.148555576781 --img_size 256 256 --init_lr 0.0001 --cluster_exams --metadata_dir /home/administrator/Mounts/Isilon/metadata --model_name resnet18 --num_chan 3 --num_workers 20 --objective cross_entropy --optimizer adam --patience 10 --pretrained_on_imagenet --run_prefix snapshot --save_dir snapshot --train --test --image_transformers scale_2d rand_hor_flip rand_ver_flip rotate_90 rotate_range/max=10/min=-10 --tensor_transformers force_num_chan_2d normalize_2d

Command to train the ImageOnly DL (MIRAI v0.1)

CUDA_VISIBLE_DEVICES=0,1,2,3 python -u scripts/main.py  --batch_size 24 --batch_splits 2 --cuda --dataset mgh_mammo_5year_risk --cluster_exams --weight_decay 5e-05 --momentum 0.9 --epochs 15 --lr_decay 0.1 --img_dir /home/administrator/Mounts/pngs16 --train_years 2012 2011 2010 2009 --dev_years 2012 2011 2010 2009 --test_years 2012 2011 2010 2009 --img_mean 7047.99 --img_size 1664 2048 --img_std 12005.5 --init_lr 0.0001 --metadata_dir /home/administrator/Mounts/Isilon/metadata --model_name custom_resnet --pool_name GlobalMaxPool --block_layout BasicBlock,2 BasicBlock,2 BasicBlock,2 BasicBlock,2 --dropout 0 --num_chan 3 --tuning_metric auc --num_workers 24 --objective cross_entropy --optimizer adam --patience 10 --max_batches_per_train_epoch 1500 --max_batches_per_dev_epoch 1500 --pretrained_on_imagenet --run_prefix snapshot --save_dir snapshot/ --train --test --class_bal --image_transformers scale_2d align_to_left rand_ver_flip rotate_range/min=-20/max=20 --tensor_transformers force_num_chan_2d normalize_2d --test_image_transformers scale_2d align_to_left --test_tensor_transformers force_num_chan_2d normalize_2d --data_parallel --num_gpus 4 --num_shards 1

Command to train the HybridDL (MIRAI v0.2)

CUDA_VISIBLE_DEVICES=4,5,6 python -u scripts/main.py  --batch_size 24 --batch_splits 2 --cuda --dataset mgh_mammo_5year_risk --cluster_exams --weight_decay 5e-05 --momentum 0.9 --epochs 15 --lr_decay 0.1 --img_dir /home/administrator/Mounts/pngs16 --train_years 2012 2011 2010 2009 --dev_years 2012 2011 2010 2009 --test_years 2012 2011 2010 2009 --img_mean 7047.99 --img_size 1664 2048 --img_std 12005.5 --init_lr 0.0001 --metadata_dir /home/administrator/Mounts/Isilon/metadata --model_name custom_resnet --pool_name GlobalMaxPool --block_layout BasicBlock,2 BasicBlock,2 BasicBlock,2 BasicBlock,2 --num_chan 3 --tuning_metric auc --num_workers 24 --objective cross_entropy --optimizer adam --patience 10 --max_batches_per_train_epoch 1500 --max_batches_per_dev_epoch 1500 --pretrained_on_imagenet --run_prefix snapshot --save_dir snapshot/ --train --test --class_bal --image_transformers scale_2d align_to_left rand_ver_flip rotate_range/min=-20/max=20 --tensor_transformers force_num_chan_2d normalize_2d --test_image_transformers scale_2d align_to_left --test_tensor_transformers force_num_chan_2d normalize_2d --data_parallel --num_gpus 3 --num_shards 1 --use_risk_factors --dropout 0 --risk_factor_metadata_path /home/administrator/Mounts/Isilon/metadata/risk_factors_aug06_2018_mammo_and_mri.json --risk_factor_keys density binary_family_history binary_biopsy_benign binary_biopsy_LCIS binary_biopsy_atypical_hyperplasia age menarche_age menopause_age first_pregnancy_age prior_hist race parous menopausal_status weight height ovarian_cancer ovarian_cancer_age ashkenazi brca mom_bc_cancer_history m_aunt_bc_cancer_history p_aunt_bc_cancer_history m_grandmother_bc_cancer_history p_grantmother_bc_cancer_history sister_bc_cancer_history mom_oc_cancer_history m_aunt_oc_cancer_history p_aunt_oc_cancer_history m_grandmother_oc_cancer_history p_grantmother_oc_cancer_history sister_oc_cancer_history hrt_type hrt_duration hrt_years_ago_stopped

About

Developing Deep Learning Models for Mammography

http://learningtocure.csail.mit.edu


Languages

Language:Python 100.0%