Periodic Reporting for period 3 - PRIMAGE (PRedictive In-silico Multiscale Analytics to support cancer personalized diaGnosis and prognosis, Empowered by imaging biomarkers)
Okres sprawozdawczy: 2021-12-01 do 2023-05-31
The digital transformation of healthcare systems has fostered the development of a vast range of innovative clinical workflows over the last decade. Particularly, the inclusion of digital diagnosis tools and patient electronic records in routine clinical practice have proven to be key enabling factors for innovation in the fields of imaging, pathology, genomic analytics and wearable sensors, among many others. In this context, real-world data-driven inferences are helping reveal insights implicit in the data, while addressing current unanswered clinical questions and generating knowledge in a more reproducible way. The development of predictive models for disease-related outcomes by learning from retrospective data, taking into account all types of clinical, pathology, molecular and imaging information, is a cornerstone of scientific progress nowadays.
The PRIMAGE project focuses on the development of in silico tools for a more personalized clinical management of cancer by targeting clinical endpoints, with two pilots focused on childhood cancer. Mathematical and computational modelling of biological processes have been used to enhance quantitative understanding of cancer progression. Since cancer patients are regularly monitored for staging and treatment response follow-up, oncologic imaging represents a suitable field for the discovery and validation of new biomarkers from different imaging modalities.
To contribute to this, the PRIMAGE project has developed an open hybrid cloud and high performance computing (HPC) platform which performance has been validated in observational studies performed on real-world data, hoping the platform will be used as a clinical decision support system in the clinical management of two paediatric malignant solid brain tumours, neuroblastoma (NB) and the Diffuse Intrinsic Pontine Glioma (DIPG). The project has utilized novel HPC approaches to support the execution of computationally efficient and large-scale in silico predictive models adressing tumor growth and response to treatment. In addition, this project has contributed to the discovery of imaging biomarkers and the development of radiomics models paving the way towards making imaging a more robust cancer diagnosis and management tool.
We expect that the methodologies used in this project for data management, in silico model development and result visualization tools can be extrapolated for their use in other cancer types impacting not only NB and DIPG but also in other malignant solid tumours.
The PRIMAGE project focuses on the development of in silico tools for a more personalized clinical management of cancer by targeting clinical endpoints, with two pilots focused on childhood cancer. Mathematical and computational modelling of biological processes have been used to enhance quantitative understanding of cancer progression. Since cancer patients are regularly monitored for staging and treatment response follow-up, oncologic imaging represents a suitable field for the discovery and validation of new biomarkers from different imaging modalities.
To contribute to this, the PRIMAGE project has developed an open hybrid cloud and high performance computing (HPC) platform which performance has been validated in observational studies performed on real-world data, hoping the platform will be used as a clinical decision support system in the clinical management of two paediatric malignant solid brain tumours, neuroblastoma (NB) and the Diffuse Intrinsic Pontine Glioma (DIPG). The project has utilized novel HPC approaches to support the execution of computationally efficient and large-scale in silico predictive models adressing tumor growth and response to treatment. In addition, this project has contributed to the discovery of imaging biomarkers and the development of radiomics models paving the way towards making imaging a more robust cancer diagnosis and management tool.
We expect that the methodologies used in this project for data management, in silico model development and result visualization tools can be extrapolated for their use in other cancer types impacting not only NB and DIPG but also in other malignant solid tumours.
Throughout the course of the project, the following main results have been achieved:
- Creation, set-up and external validation of the PRIMAGE platform, currently including a final number of cases of 1148 for NB and 71 cases for DIPG. From these cases, several predictive models have been developed and integrated in the final version of the platform as explained below. The platform incorporates a set of visual analytics tools, which together with the whole platform have been validated by external collaborators and clinical partners of the consortium. Following this validation, the PRIMAGE platform final user interface has been adapted to improve usability and user experience.
- Development and integration in the platform of several imaging biomarkers extraction pipelines for both NB and DIPG, including radiomics features for T2W, T1W and CT, deep features for T2W, T1W and CT, Diffusion ADC maps, Fit-Cluster-Fit and semiquantitative perfusion analyses. In addition, NB and DIPG segmentation models have been trained and validated, which has enabled automatic tumour delineation for the automation of the imaging biomarker extraction. Machine Learning models with clinical, molecular and imaging variables have been developed for the prediction of overall survival and event free survival in NB.
- Development of an orchestrated multiscale tumor growth model for NB tumors that enables clinicians to launch simulations.
- Development of AI-based models to generate responses to the most relevant Clinical End Point for NB, particularly overall survival (OS) and event-free survival (EFS). Based on this information, a diagnostic module for NB patient stratification has been integrated in the platform. This module closes the loop and proves the capacity of the PRIMAGE platform to be used as a clinical decision support system, as the main results obtained by this module can be visualized in a pdf report that has been specifically designed to increase trust in AI models.
All together, the work performed in the Primage project has led to the publication of 38 scientific publications (with 6+ currently in writing) and has created two main exploitable assets: the Primage platform and the data base.
- Creation, set-up and external validation of the PRIMAGE platform, currently including a final number of cases of 1148 for NB and 71 cases for DIPG. From these cases, several predictive models have been developed and integrated in the final version of the platform as explained below. The platform incorporates a set of visual analytics tools, which together with the whole platform have been validated by external collaborators and clinical partners of the consortium. Following this validation, the PRIMAGE platform final user interface has been adapted to improve usability and user experience.
- Development and integration in the platform of several imaging biomarkers extraction pipelines for both NB and DIPG, including radiomics features for T2W, T1W and CT, deep features for T2W, T1W and CT, Diffusion ADC maps, Fit-Cluster-Fit and semiquantitative perfusion analyses. In addition, NB and DIPG segmentation models have been trained and validated, which has enabled automatic tumour delineation for the automation of the imaging biomarker extraction. Machine Learning models with clinical, molecular and imaging variables have been developed for the prediction of overall survival and event free survival in NB.
- Development of an orchestrated multiscale tumor growth model for NB tumors that enables clinicians to launch simulations.
- Development of AI-based models to generate responses to the most relevant Clinical End Point for NB, particularly overall survival (OS) and event-free survival (EFS). Based on this information, a diagnostic module for NB patient stratification has been integrated in the platform. This module closes the loop and proves the capacity of the PRIMAGE platform to be used as a clinical decision support system, as the main results obtained by this module can be visualized in a pdf report that has been specifically designed to increase trust in AI models.
All together, the work performed in the Primage project has led to the publication of 38 scientific publications (with 6+ currently in writing) and has created two main exploitable assets: the Primage platform and the data base.
PRIMAGE faced different technological challenges during its development, which have made a great impact in the translation of big & multi-disciplinary data into predictors for personalised decision making; the improvement of data quality and interoperability standards,. Some of these challenges implied the development of:
- Open cloud infrastructures for healthcare, ensuring privacy management, transparency, liability and long-term sustainability.
- A Hybrid Cloud Model combining restricted sites within healthcare networks processing the sensitive data and external clouds that only deal with anonymised data.
- Reproducible imaging data in a standardized format to provide large sample size statistics and generalizable trends.
- A multi-step image preprocessing pipeline for image preparation addressing the various challenges of medical image analysis.
- Automatic segmentation algorithms based on encoder-decoder CNNs and the addition of multi-sequence information.
- A multiscale hybrid approach that combines both stochastic and continuum strategies in order to simulate NB and DIPG tumour growth, vascularization, residual stresses and extracellular matrix stiffening.
- A DSS offering visual analytics and predictive tools to assist management of NB and DIPG paediatric cancers based on the use of novel imaging biomarkers, tumour growth models and advanced visualisation of predictions.
- Open cloud infrastructures for healthcare, ensuring privacy management, transparency, liability and long-term sustainability.
- A Hybrid Cloud Model combining restricted sites within healthcare networks processing the sensitive data and external clouds that only deal with anonymised data.
- Reproducible imaging data in a standardized format to provide large sample size statistics and generalizable trends.
- A multi-step image preprocessing pipeline for image preparation addressing the various challenges of medical image analysis.
- Automatic segmentation algorithms based on encoder-decoder CNNs and the addition of multi-sequence information.
- A multiscale hybrid approach that combines both stochastic and continuum strategies in order to simulate NB and DIPG tumour growth, vascularization, residual stresses and extracellular matrix stiffening.
- A DSS offering visual analytics and predictive tools to assist management of NB and DIPG paediatric cancers based on the use of novel imaging biomarkers, tumour growth models and advanced visualisation of predictions.