Main Authors: Luis Martí-Bonmatí (HUPLF), Ángel Alberich-Bayarri (QUBIM), Adela Cañete (HUPLF)

Adittional Authors: Ignacio Blanquer (UPV), Marian Bubak (CYFRONET), Mario Aznar (MATICAL), Gracia Marti (HUPLF)

Focus Area: 

Cloud Computing

Who stands to benefit and how: 

Medical data scientist facing computational and storage issues when processing their data.

Position Paper: 

Computational imaging allows the extraction of multiparametric data, leading to a new era in Radiomics (the extraction of disease features from medical images using data-characterisation and modelling algorithms), supported by the high-throughput extraction, storage, and analysis of a large amount of quantitative imaging features with clinical surrogates (Imaging Biomarkers), providing quantitative information (Virtual Biopsies) for early disease diagnosis, phenotyping, grading, targeting therapies and evaluation of disease response to treatment. Oncologic imaging represents a suitable field for the discovery and validation of new biomarkers from different imaging modalities since oncologic patients are frequently monitored for staging and follow-up of treatment response.
Many imaging biomarkers have been proposed in the last years to measure tumour semantics, pathophysiology, metabolism and molecular profiles in order to estimate different cancer hallmarks, such as proliferation/growth, angiogenesis, evasion or metastasis. However, a very limited number of biomarkers have entered into routine clinical practise to guide clinical decisions, as yet. The majority of oncology imaging biomarkers still need external validation at different centres before they can be properly qualified as robust biomarkers.

PRIMAGE is one of the largest and more ambitious research projects dealing with medical imaging, artificial intelligence and cancer treatment. This four-year European Commission funded project has 16 European partners in the consortium, including the European Society for Paediatric Oncology, two Imaging Biobanks and three prominent European Paediatric oncology units. This observational in the silico study involves high-quality retrospective anonymized data (imaging, clinical, molecular and genetics) for the training of machine learning and multiscale algorithms.
PRIMAGE will develop predictive models using artificial intelligence to predict valid disease outcomes by learning on retrospective data is a hot topic debate. The validity of these predictive models depends on the quantity, quality and representativeness of the datasets used, being a major limiting factor. PRIMAGE will also develop in silico models requiring significant computational and data storage resources to process.

The foundation back-end of PRIMAGE will combine large-scale high-performance computing (HPC) and versatile cloud computing resources for optimum efficiency and reliability. For this purpose PRIMAGE will gather:

  • Large-scale processing on HPC resources, overlaid by a convenient REST-based process controller called Rimrock and data access suite.
  • Hybrid cloud resources, composed of both private and public cloud sites (based on EOSC services), which will host PRIMAGE data repositories. Computational tasks can be deployed and coherently managed by a single access tool, such as the ATMOSPHERE orchestrator.
  • An integration middleware, consisting of a set of protocols and interfaces between high-performance computing/storage (models), private and public cloud computing/storage (repositories, sandboxed processing) and external data sources (anonymised clinical and biobanking data). The middleware will be put in place to achieve an adequate level of solution coherency.
  • Upper layer User Interface service exposing the features of the underlying infrastructure to researchers as a convenient Graphical User Interface, to manage definition, execution and comparison of results of computationally-intensive model pipelines. The tool will feature security management and Application Programming Interface for programmatic access, and it will be based on the Model Execution Environment.

The open cloud-based platform will offer assistance for phenotyping (diagnosis), treatment allocation (prediction) and patient clinical endpoints (prognosis), based on the use of imaging biomarkers, tumour growth simulation, advanced visualisation of confidence scores, and machine-learning approaches. The decision support prototype will be constructed and validated on two paediatric cancers: neuroblastoma and diffuse intrinsic pontine glioma. The results will be available for the scientific community and ready for transfer learning to other malignant solid tumours. Data infrastructures, imaging biomarkers and predictive models for in silico medicine research will be validated during this project in neuroblastoma (NB) and the diffuse intrinsic pontine glioma (DIPG) context. NB is the most frequent solid cancer of early childhood. The age at diagnosis has proven to be crucial factor in its prognosis. DIPG is the leading cause of brain tumor-related death in children. Given the rarity of NB and DIPG, international cooperative networks are essential to agglutinate relevant retrospective data and/or prospective cases for clinical trials, facilitating the identification of effective tools for earlier diagnosis and potentially effective therapeutics.