The main goal of the DeepHealth project is to put HPC computing power at the service of biomedical applications; and apply Deep Learning (DL) and Computer Vision (CV) tech-niques on large and complex biomedical datasets to support new and more efficient ways of diagnosis, monitoring and treatment of diseases. 

THE DEEPHEALTH TOOLKIT: A KEY OPEN-SOURCE ASSET FOR EHEALTH AI-BASED SOLUTIONS 

At the centre of the proposed innovations is the DeepHealth toolkit, an open-source free software that will provide a unified frame-work to exploit heterogeneous HPC and big data architectures assembled with DL and CV capabilities to optimise the training of predictive models. The toolkit is composed of two core libraries plus a back end offering a REST-ful API to other applications to facilitate the use of the libraries, and a dedicated front end that interacts with the back end for facilitating the use of the libraries to data scientists. The libraries are the European Distributed Deep Learning Library (EDDL) and the Euro-pean Computer Vision Library (ECVL). Version 1.0 of both libraries has been released during second semester of 2021, and both are ready to be integrated in current and new biomedical platforms or applications. 

THE DEEPHEALTH CONCEPT – APPLICATION SCENARIOS 

The DeepHealth concept focuses on scenarios where image processing is needed for diagnosis. In the training environment IT experts work with datasets of images for training predictive models. In the production environment the medical personnel ingests an image coming from a scan into a platform or biomedical application that uses predictive models to get clues to support them during diagnosis. The DeepHealth toolkit will allow the IT staff to train models and run the train-ing algorithms over hybrid HPC + big data architectures without a profound knowledge of DL, CV, HPC or big data and increase their productivity reducing the required time to do it. 

14 PILOTS AND SEVEN PLATFORMS TO VALIDATE THE DEEPHEALTH PROPOSED INNOVATIONS 

The DeepHealth innovations are being validated in 14 pilots through the use of seven different biomedical and AI software plat-forms provided by partners. The libraries are being integrated and validated in seven AI and biomedical software platforms: commercial platforms: (everis Lumen, PHILIPS Open Innovation Platform, THALES PIAF) and research-oriented platforms (CEA’s ExpressIFTM, CRS4’s Digital Pathology, WINGS MigraineNet). 

The use cases cover three main areas: (i) Neurological diseases, (ii) Tumour detection and early cancer prediction and (iii) Digital pathology and automated image annotation. The pilots will allow evaluating the performance of the proposed solutions in terms of the time needed for pre-processing images, the time needed to train models and the time to put models in production. In some cases, it is expected to reduce these times from days or weeks to just hours. This is one of the major expected impacts of the DeepHealth project.