Welcome to the 6th DeepHealth newsletter! Throughout these communications, we provide more relevant news about the project.
The DeepHealth project officially ended in July, and the final review is expected in September. We want to thank all people following us and give a hint of the leading project outcomes and follow-up, starting by illustrating the Success Stories that demonstrate how clinical people can benefit from the DeepHealth technology.
Success Stories
Success Stories leverage on the DeepHealth pilots and tell how each used technologies and/or platforms developed in the project to solve a specific health problem, focusing on how doctors and medical personnel can benefit from this approach. Take a look at the DeepHealth website and find more about:
The DeepHealth Toolkit is a general-purpose deep learning framework, including image processing and computer vision functionalities, enabled to exploit HPC and cloud infrastructures for running parallel/distributed training and inference processes. The core of the toolkit consists of two libraries, namely:
The European Distributed Deep Learning Library (EDDL)
The European Computer Vision Library (ECVL)
The Front-end and Back-end accompany the libraries to allow and facilitate their use. The libraries are designed to run under the control of HPC-specific workflow managers such as COMPSs and StreamFlow. All the toolkit components are available as free and open-source software under the MIT license.
The EDDL is an optimized C++ library for Distributed Deep Learning and Tensor Operations. It allows the definition of different neural network architectures and their execution on different hardware platforms and accelerators, offering tailored implementations on CPUs, Graphical Processing Units (GPUs), and Field Programmable Gate Arrays (FPGA). A python wrapper PyEDDL is also available.
The ECVL is an image processing software stack that supports the most common image formats (medical & non-medical) and whole-slide images. The ECVL library has also been adapted to be used in a cloud environment by extending it to perform parallel and distributed jobs on Kubernetes-managed computing resources. A python wrapper PyECVL is also available.
The DeepHealth toolkit Front-end is a web-based graphical, intuitive and attractive user interface that allows the user to access all of the functionalities provided by both libraries.
The DeepHealth Back-end manages low-level libraries interaction, driving the dataset loading, image manipulation, and the training and inference processes; through the front-end part, the user exploits these functionalities in an easy-to-use way.
COMP Superscalar (COMPSs) is a programming model developed by BSC that aims to ease application development for distributed infrastructures. In DeepHealth, it is used to efficiently distribute the computation of the libraries while hiding the infrastructure complexities.
StreamFlow is a workflow management system developed by UNITO to support the execution of hybrid workflows. It adopts the Common Workflow Language (CWL) open standard to express steps and dependencies, relying on platform-dependent description formats to specify locations in charge of executing them.
DeepHealth Platforms
The DeepHealth project offers an extraordinary panel of platforms showing the adaptability and flexibility of the DeepHealth Libraries. The DeepHealth core technology was integrated successfully to:
A mobile application platform, Migraine Net (WINGS), allowing direct patient follow-up via mobile device
Cloud platforms, such as Lumen (NTT Data), OpenDeepHealth (UNITO), and Digital Pathology (CRS4), increasing the efficiency of the health industry and making AI applications safe
Web platforms, such as Piaf (Thales) and Open Innovation (PHILIPS), allowing fast deployment for testing and inference of AI algorithms on new datasets
Merge rule-based engine, as demonstrated by ExpressIF (CEA)
