This code is for eCaReNet: explainable Cancer Relapse Prediction Network.

Related tags

Deep Learningecarenet
Overview

eCaReNet

This code is for eCaReNet: explainable Cancer Relapse Prediction Network. (Towards Explainable End-to-End Prostate Cancer Relapse Prediction from H&E Images Combining Self-Attention MultipleInstance Learning with a Recurrent Neural Network, Dietrich, E., Fuhlert, P., Ernst, A., Sauter, G., Lennartz, M., Stiehl, H. S., Zimmermann, M., Bonn, S. - ML4H 2021)

eCaReNet takes histopathology images (TMA spots) as input and predicts a survival curve and a risk score for individual patients. The network consists of an optional self-attention layer, an RNN and an attention based Mulitple Instance Learning module for explainability. To increase model performance, we suggest to include a binary prediction of a relapse as input to the model. alt text

TL;DR

  • store your dataset information in a .csv file

  • make your own my_config.yaml, following the example in config.yaml

  • run $ python train_model.py with my_config.yaml

Requirements and Installation

  • Python and Tensorflow
  • npm install -g omniboard to view results in browser and inspect experiments

Data preprocessing

All annotations of your images need to be stored in a .csv file with the image path and annotations as columns. You need separate csv files for your training, validation and test sets. Here is an example:

img_path censored relapse_time survived_2years ISUP_score
img1.png 0 80.3 1 3

The columns can be named as you wish, you need to tell python which columns to use in the config file ↓

config file

The config file (config.yaml) is needed to define the directories where the images and training, validation and test .csv files are stored. Further, you can choose whether to train a classification (for M_ISUP or M_Bin) or the survival model eCaReNet, which loss function and optimizer to use. Also the preprocessing is defined here (patching, resizing, ...) Details are found in config.yaml. It is best to create a custom my_config.yaml file and run the code as

$ python train_model.py with my_config.yaml 

You can also change single parameters in the command line like

$ python train_model.py with my_config.yaml general.seed="13" 

Training procedure

As recommended in our paper, we suggest to first train M_ISUP with config_isup.yaml. So train your base network (Inception or other) on a classification task as a transfer learning task. Second, train a binary classifier with config_bin.yaml, choose an appropriate time point to base the decision on. Here, you need to load the pretrained model from step one, do not load Inception or other keras models. For the third step the prediction from model two M_Bin are needed, so please store the information in the .csv file. Then again, load model from step one, and this time include the predictions as additional input and train.

Unittests

For most functions, a unittest is given in the test folder. This can be used to test if the function works correctly after adapting it (e.g. add more functionality or speed up). Further, it can be used for debugging to find errors or to find out what the function is actually doing. This is faster than running the whole code.

Docker

In the docker_context folder, the Dockerfile and requirements are stored. To build a docker image run

$ "docker build -t IMAGE_NAME:DATE docker_context"
$ docker build  -t ecarenet_docker:2021_10 docker_context

To use the image, run something like the following. Needs to be adapted to your own paths and resources

$ docker run --gpus=all --cpuset-cpus=5,6 --user `id --user`:`id --group` -it --rm -v /home/UNAME/PycharmProjects/ecarenet:/opt/project -v /PATH/TO/DATA:/data --network NETWORK_NAME--name MY_DOCKER_CONTAINER ecarenet_docker:2021_10 bash

More information on docker can be found here: https://docs.docker.com/get-started/

sacred

We use sacred (https://sacred.readthedocs.io/en/stable/) and a MongoDB (https://sacred.readthedocs.io/en/stable/observers.html https://www.mongodb.com/) to store all experiments. For each training run, a folder with an increasing id will be created automatically and all information about the run will be stored in that folder, like resulting weights, plots, metrics and losses. Which folder this is, is written in settings/default_settings and in the config in training.model_save_path.

The code works without mongodb, if you want. All results will be stored. If you do want to use the mongodb, you need to run a docker container with a mongoDB:

$ docker run -d -p PORTNUMBER:27017 -v ./my_data_folder:/data/db --name name_of_my_mongodb mongo

Create a network:

$ docker network create NETWORK_NAME

Attach container to network

$ docker network connect NETWORK_NAME name_of_my_mongodb

Then during training, the --network NETWORK_NAME property needs to be set. Use omniboard to inspect results:

$ omniboard -m localhost:PORTNUMBER:sacred

Tensorflow

Using tensorflow tf.data speeds up the data generation and preprocessing steps. If the dataset is very large, it can be cached with tf.cache() either in memory or to a file with tf.cache('/path/to/folder/plus/filename') [in dataset_creation/dataset_main.py]. Using tensorflow, it is also best to not overly use numpy functions or to decorate them with tf.function. Functions decorated with @tf.function will be included in the tensorflow graph in the first step and not be created again and again. For debugging, you need to remove the tf.function decorator, because otherwise the function (and breakpoints inside) will be skipped.

Owner
Institute of Medical Systems Biology
Institute of Medical Systems Biology
Code for the paper "PortraitNet: Real-time portrait segmentation network for mobile device" @ CAD&Graphics2019

PortraitNet Code for the paper "PortraitNet: Real-time portrait segmentation network for mobile device". @ CAD&Graphics 2019 Introduction We propose a

265 Dec 01, 2022
Semantic Segmentation Suite in TensorFlow

Semantic Segmentation Suite in TensorFlow. Implement, train, and test new Semantic Segmentation models easily!

George Seif 2.5k Jan 06, 2023
Code for our paper "SimCLS: A Simple Framework for Contrastive Learning of Abstractive Summarization", ACL 2021

SimCLS Code for our paper: "SimCLS: A Simple Framework for Contrastive Learning of Abstractive Summarization", ACL 2021 1. How to Install Requirements

Yixin Liu 150 Dec 12, 2022
OOD Dataset Curator and Benchmark for AI-aided Drug Discovery

🔥 DrugOOD 🔥 : OOD Dataset Curator and Benchmark for AI Aided Drug Discovery This is the official implementation of the DrugOOD project, this is the

108 Dec 17, 2022
yolox_backbone is a deep-learning library and is a collection of YOLOX Backbone models.

YOLOX-Backbone yolox-backbone is a deep-learning library and is a collection of YOLOX backbone models. Install pip install yolox-backbone Load a Pret

Yonghye Kwon 21 Dec 28, 2022
Python implementation of cover trees, near-drop-in replacement for scipy.spatial.kdtree

This is a Python implementation of cover trees, a data structure for finding nearest neighbors in a general metric space (e.g., a 3D box with periodic

Patrick Varilly 28 Nov 25, 2022
Single Red Blood Cell Hydrodynamic Traps Via the Generative Design

Rbc-traps-generative-design - The generative design for single red clood cell hydrodynamic traps using GEFEST framework

Natural Systems Simulation Lab 4 Jun 16, 2022
An Efficient Training Approach for Very Large Scale Face Recognition or F²C for simplicity.

Fast Face Classification (F²C) This is the code of our paper An Efficient Training Approach for Very Large Scale Face Recognition or F²C for simplicit

33 Jun 27, 2021
"Structure-Augmented Text Representation Learning for Efficient Knowledge Graph Completion"(WWW 2021)

STAR_KGC This repo contains the source code of the paper accepted by WWW'2021. "Structure-Augmented Text Representation Learning for Efficient Knowled

Bo Wang 60 Dec 26, 2022
Monocular Depth Estimation - Weighted-average prediction from multiple pre-trained depth estimation models

merged_depth runs (1) AdaBins, (2) DiverseDepth, (3) MiDaS, (4) SGDepth, and (5) Monodepth2, and calculates a weighted-average per-pixel absolute dept

Pranav 39 Nov 21, 2022
A PyTorch implementation of "Signed Graph Convolutional Network" (ICDM 2018).

SGCN ⠀ A PyTorch implementation of Signed Graph Convolutional Network (ICDM 2018). Abstract Due to the fact much of today's data can be represented as

Benedek Rozemberczki 251 Nov 30, 2022
Implementation of CVPR 2021 paper "Spatially-invariant Style-codes Controlled Makeup Transfer"

SCGAN Implementation of CVPR 2021 paper "Spatially-invariant Style-codes Controlled Makeup Transfer" Prepare The pre-trained model is avaiable at http

118 Dec 12, 2022
Gym environment for FLIPIT: The Game of "Stealthy Takeover"

gym-flipit Gym environment for FLIPIT: The Game of "Stealthy Takeover" invented by Marten van Dijk, Ari Juels, Alina Oprea, and Ronald L. Rivest. Desi

Lisa Oakley 2 Dec 15, 2021
PyTorch implementation code for the paper MixCo: Mix-up Contrastive Learning for Visual Representation

How to Reproduce our Results This repository contains PyTorch implementation code for the paper MixCo: Mix-up Contrastive Learning for Visual Represen

opcrisis 46 Dec 15, 2022
Official pytorch code for "APP: Anytime Progressive Pruning"

APP: Anytime Progressive Pruning Diganta Misra1,2,3, Bharat Runwal2,4, Tianlong Chen5, Zhangyang Wang5, Irina Rish1,3 1 Mila - Quebec AI Institute,2 L

Landskape AI 12 Nov 22, 2022
Official PyTorch implementation of RobustNet (CVPR 2021 Oral)

RobustNet (CVPR 2021 Oral): Official Project Webpage Codes and pretrained models will be released soon. This repository provides the official PyTorch

Sungha Choi 173 Dec 21, 2022
Human pose estimation from video plays a critical role in various applications such as quantifying physical exercises, sign language recognition, and full-body gesture control.

Pose Detection Project Description: Human pose estimation from video plays a critical role in various applications such as quantifying physical exerci

Hassan Shahzad 2 Jan 17, 2022
Supervised Sliding Window Smoothing Loss Function Based on MS-TCN for Video Segmentation

SSWS-loss_function_based_on_MS-TCN Supervised Sliding Window Smoothing Loss Function Based on MS-TCN for Video Segmentation Supervised Sliding Window

3 Aug 03, 2022
PAIRED in PyTorch 🔥

PAIRED This codebase provides a PyTorch implementation of Protagonist Antagonist Induced Regret Environment Design (PAIRED), which was first introduce

UCL DARK Lab 46 Dec 12, 2022
Exact Pareto Optimal solutions for preference based Multi-Objective Optimization

Exact Pareto Optimal solutions for preference based Multi-Objective Optimization

Debabrata Mahapatra 40 Dec 24, 2022