Pediatric Bone Age Assessment

In this project, we introduce the problem of pediatric bone age assessment. During an organism’s development, the bones of the skeleton change in size and shape. Difference between a child’s assigned bone age and chronological age might indicate a growth problem. Clinicians use bone age assessment in order to estimate the maturity of a child’s skeletal system.

Bone age assessment usually starts with taking a single X-ray image of the left hand from wrist to fingertips. Traditionally, bones in the radiograph are compared with images in a standardized atlas of bone development. This recipe represents a core approach described in "Paediatric Bone Age Assessment Using Deep Convolutional Neural Networks" by V. Iglovikov, A. Rakhlin, A. Kalinin and A. Shvets, link 1, 2.

We validate the performance of the method by using data from the 2017 Pediatric Bone Age Challenge organized by the Radiological Society of North America (RSNA). The data set has been contributed by 3 medical centers at Stanford University, the University of Colorado and the University of California - Los Angeles. Originally, the dataset was shared by the AIMI Center of Stanford University and now can be freely accessed at Kaggle platform. For the sake of simplicity, we skip intense preprocessing steps as described in the original work and provide radiographs with already removed background and uniformly registered hand imagess.

Original and preprocessed radiographs of a hand of 82 month old (approx. 7 y.o.) girl

Technologies

• Catalyst as pipeline runner for deep learning tasks. This new and rapidly developing library can significantly reduce the amount of boilerplate code. If you are familiar with the TensorFlow ecosystem, you can think of Catalyst as Keras for PyTorch. This framework is integrated with logging systems such as the well-known TensorBoard and the new Weights & biases.

Quick Start

0. Sign up at neu.ro

1. Install CLI and log in

pip install -U neuromation
neuro login

2. Run the recipe

git clone git@github.com:neuromation/ml-recipe-bone-age.git
cd ml-recipe-bone-age
make setup
make jupyter

3. Train the model

Download the dataset from within the demo notebook, then run:

make training

Development Environment

This project is designed to run on Neuro Platform, so you can jump into problem-solving right away.

Directory structure

Local directory	Description	Storage URI	Environment mounting point
data/	Data	storage:ml-recipe-bone-age/data/	/ml-recipe-bone-age/data/
src/	Python modules	storage:ml-recipe-bone-age/src/	/ml-recipe-bone-age/src/
notebooks/	Jupyter notebooks	storage:ml-recipe-bone-age/notebooks/	/ml-recipe-bone-age/notebooks/
No directory	Logs and results	storage:ml-recipe-bone-age/results/	/ml-recipe-bone-age/results/

Development

Follow the instructions below to set up the environment and start your Jupyter Notebook development session.

Setup development environment

make setup

• Several files from the local project upload to the platform’s storage (namely, requirements.txt, apt.txt, setup.cfg).
• A new job starts in our base environment.
• Pip requirements from requirements.txt and apt applications from apt.txt install in this environment.
• The updated environment is saved under a new project-dependent name and will be used later on.

Run Jupyter with GPU

make jupyter

• The content of modules and notebook directories upload to the platform’s storage.
• A job with Jupyter is started, and its web interface opens in the local web browser window.

Kill Jupyter

make kill-jupyter

This command terminates the job with Jupyter Notebooks. The notebooks remain saved on the platform’s storage. If you’d like to download them to the local notebooks/ directory, just run make download-notebooks.

Help

make help

Data

Uploading via Web UI

On your local machine, run make filebrowser and open the job's URL on your mobile device or desktop. Through a simple file explorer interface, you can upload test images and perform file operations.

Uploading via CLI

On your local machine, run make upload-data. This command pushes local files from ./data into storage:ml-recipe-bone-age/data and mounts them to your development environment's /project/data.

Customization

Several variables in Makefile are intended to be modified according to the project’s specifics. To change them, find the corresponding line in Makefile and update it.

Data location

DATA_DIR_STORAGE?=$(PROJECT_PATH_STORAGE)/$(DATA_DIR)

This project template implies that your data is stored alongside the project. If this is the case, you don't need to change this variable. However, if your data is shared between several projects on the platform, you will need to change the following line to point to its location. For example:

DATA_DIR_STORAGE?=storage:datasets/cifar10

Training machine type

TRAINING_MACHINE_TYPE?=gpu-small

There are several machine types supported on the platform. Run neuro config show to see the list.

HTTP authentication

HTTP_AUTH?=--http-auth

When jobs with HTTP interface are executed (for example, with Jupyter Notebooks or TensorBoard), this interface requires that the user be authenticated on the platform. However, if you want to share the link with someone who is not registered on the platform, you may disable the authentication requirement by updating this line to HTTP_AUTH?=--no-http-auth.

Training command

TRAINING_COMMAND?='echo "Replace this placeholder with a training script execution"'

If you want to train some models from code instead of from Jupyter Notebook, you need to update this line. For example:

TRAINING_COMMAND="bash -c 'cd $(PROJECT_PATH_ENV) && python -u $(CODE_DIR)/train.py --data $(DATA_DIR)'"