Skip to main content

Big Data and models for personalized Head and Neck Cancer Decision support

Periodic Reporting for period 2 - BD2Decide (Big Data and models for personalized Head and Neck Cancer Decision support)

Reporting period: 2017-12-01 to 2018-11-30

Cancers of the Head and Neck region (HNC) are the 6th most deadly tumors worldwide, with ~630.000 newly detected cases (150.000 in Europe) and ~350.000 deaths every year (~70,000 in Europe). These cancers are increasing especially in the younger population and constitute a significant cause of mortality. Approximately two third of HNC are detected at advanced stage (Stage III and IV); their treatment requires multimodal interventions, with a curability rate of approximately 60%. 27%-50% of cases usually relapse within two years after treatment and salvage surgery or re-irradiation are ineffective in the majority of cases. Treatment of HNC is extremely invasive and impairing causing serious side effects to patients; treatment selection is a critical phase of patients' management. Currently patients treatment is decided based on the Tumor-Lymph-nodes-Metastases system, which considers only a few prognostic factors, not sufficient to personalize treatment. Physicians need a more accurate stratification of patients according to prognostic biomarkers able to predict the individual patient’s clinical outcome at the time diagnosis.
BD2Decide realizes and validates a Clinical Decision Support System linking together population-specific epidemiology, behavioral and environmental data, patient-specific multiscale data from genomics, pathology, clinical and imaging data with available multiscale prognostic models and Graphical Visualization tools that allow to:
1. Improve the clinical decision process, by implementing a model-based prognostic system, expected to increase the accuracy of current TNM staging by 10%.
2. Validate the prognostic models in different populations.
3. Uncover patient-specific and population-related patterns that can improve care and patients' quality of life, and pave the way for better tailored treatment guidelines.
4. Reinforce the multidisciplinary decision-making process and patient's co-decision through advanced data visualization and presentation.
5. Create a virtuous circle of learning between research and clinical practice, through the the creation of a large and shared data repository for HNC.
The clinical study approved by ethics committees and registered in clinicaltrials.gov of participating hospitals is progressing. A total of 1087 retrospective and 389 prospective patients have been enrolled for whom clinical/medical, pathology, treatment, toxicity and follow-up data have been collected. Images for >833 retrospective and 178 prospective patients have been collected, processed and radiomics data extracted; biologic specimens for >700 retrospective and >150 prospective patients have been collected and are being processed and analyzed; patients selection and data collection completion and data quality check is ongoing. Population data (cancer registries for 5 EU Countries and socio-economic, lifestyle, environmental data) have been collected and used for model recalibration andbig data analysis. HPV study on oropharynx (N=1200) oral cavity (N>1000) cancer cases was conducted at VUmc and preliminary results are available.
Use scenarios have been defined, the system architecture consolidated and the BD2Decide modules realized, tested and integrated in the Big Data Infrastructure; local data repositories are established at each clinical partners' site in order to maintain the original patients' data locally. Data security and protection are in place. Fraunhofer released the complete Image Analysis Tool Suite for the semi-automatic segmentation of tumor and lymph-nodes, the extraction of morphological features from CTs and MRIs. POLIMI and MAASTRO radiomic software tools have been tested, radiomic features extraction and signatures validation is ongoing. Prognostic models are being recalibrated and evaluated. In WP5 the Patients Documentation System has been realized, and the visualization and data presentation prototype tools have been completed and validated by end users. MAASTRO has completed the Interactive Patients co-Decision aid tool for larynx cancer in Dutch and English language; translation into German is ongoing. UPM developed the BD2Decide ontology and the Visual Analytics Tools for data query and simulations. The Big Data Infrastructure and a first prototype of Big Data Analysis tools, providing first results that are now validated by users. The technical validation is in advanced stage and the health technology assessment has started. Dissemination activities in the period allowed to reach more than 5.900 researchers in the scientific community, more than 250 industrial representatives, 1.860 representatives of civil society, 56 policy makers, 10 investors, 24 media, and more than 50.000 citizens in the general public. The project logo, website and presence in social media was established in the first months of the project. From the project start the website had a 2885 visits from 2569 users. The prospect market has been sized, IPRs have been defined and preliminary exploitation intentions produced. 34 official deliverables were submitted.
"The main innovations brought by BD2Decide vs. existing clinical decision support systems can be summarized in five main achievements:
1. The introduction of radiomics into the diagnostic and treatment decision process and into prognostic models.
2. The integration of different models, their recalibration considering all the multiscale factors involved in HNC progression and on population data.
3. The application of cloud computing and big data analysis techniques to uncover unexpected prognostic indicators.
4. The introduction of new collaborative concepts in patients' management: collaborative decision making through data sharing, prognostic modeling and data investigation.
5. The implementation of a ""transparent"" data presentation and prognostic modeling output.
The major results expected from the Project consist in the implementation of the following components:
• Big Data Infrastructure
• Big Data Analytics and Visual Analytics Tool.
• Clinical Decision Support System Tool and Knowledge Management System.
• BD2Decide Ontology.
• Library of Statistical models.
• Images segmentation tool, CT and MRI radiomic feature extractor tools and Phenotipization tool to extract radiomic signatures from MRIs.
• Interactive Patients' co-Decision Aid.

BD2Decide work entails relevant impact in several directions:
• Improve the clinical decision process for patients diagnosed with HNC and by creating a virtuous circle of learning between research and clinical practice, through the mutual feeding of multiple data categories.
• Demonstrate and exploit the value potential of big data for prognostic prediction, to discovery personalized prognostic patterns for HNC that outperform the currently used TNM staging system
• By improving the clinical decision making process and by involving the patients in such process, BD2Decide will have crucial impact on the way trade-off between clinical effectiveness and patients Quality of Life (QoL) is currently managed.
• Through more accurate and personalized prognosis BD2Decide is expected to contribute reduce treatment costs of HNC, which currently requires several tens of thousands of Euros per patient. This is achieved by supporting the selection of the best possible treatment options – consistent with up-to-date clinical knowledge on one side, and patients’ preferences on the other side – and by avoiding unnecessary care.
"
BD2Decide MRI radiomic tools
BD2Decide radiomics workflow
BD2decide architecture
BD2Decide prognostic modelling and Big Data analysis presentation
BD2Decide data corpus generation
BD2Decide CT radiomics signature
BD2Decide workflow
BD2Decide Visual Analytics Tools
BD2Decide BIg Data Infrastructure
BD2Decide Big Data Analysis
BD2Decide Image Analysis Tools
BD2decide approach