What is DeepCell?

DeepCell is a collection of machine learning resources that facilitate the development and application of new deep learning methods to biology by addressing 3 needs: (1) Data Annotation and Management, (2) Model Development, and (3) Deployment and Inference

Data Annotation and Management

DeepCell Label is our training data curation tool. It provides an inutitive UI for users to create annotations from scratch or to correct model predictions, to faciliate the creation of large, high-quality datasets. DeepCell Label can be deployed locally or on the cloud.

Model Development

deepcell-tf is our core deep learning library. Based on TensorFlow, it contains a suite of tools for building and training deep learning models. The library has been constructed in a modular fashion to make it easy to mix and match different model architectures, prediction tasks, and post-processing functions. For more information, check out the documentation.

Deployment & Inference

The kiosk-console is a turn-key cloud-based solution for deploying a scalable inference platform. The platform includes a simple drag-and-drop interface for segmenting a few images, and a robust API capable of affordably processing millions of images.

We use this platform to host DeepCell.org and currently deployed models. However, it is built with extensibility in mind, and it is easy to deploy your own models. To learn more about deploying your own instance of DeepCell.org using the kiosk-console, read the docs.

The kiosk-imagej-plugin enables ImageJ to segment images with a deployed DeepCell Kiosk model without leaving the application.

deepcell-applications contains a variety of trained deep learning models and post-processing functions for instance segmentation. Each model can be imported and run locally from a Docker image, Jupyter notebook, or custom script.

The ark repository is our integrated multiplex image analysis pipeline. The input is multiplexed image data from any platform. It segments the data with Mesmer using the Kiosk, extracts the counts of each marker in each cell, normalizes the data, and then creates a summary table with morphological information and marker intensity for every cell in each image. It also provides an easy way to run some standard spatial analysis functions on your data.