SPAIN

Population Imaging Valencia

The Biomedical Imaging Research Group (GIBI230) at La Fe Health Research Institute (IIS La Fe) comprises a diverse team of 39 contracted researchers and 14 associated collaborators, establishing itself as a multidisciplinary clinical and technical unit centered in medical imaging developments. Proficient in computational solutions, the group specializes in extracting robust and reproducible imaging biomarkers through the utilization of radiomic methods and artificial intelligence algorithms, primarily for enhancing observational studies.

Strategically, the group focuses on developing and validating predictive models while concurrently creating extensive, structured, and standardized image repositories containing associated clinical, pathological, and molecular information (Real World Data). Drawing on considerable experience from previous European projects, the group excels in cancer imaging, distributed computing, data interoperability standards, privacy-preserving synthetic data generation, and AI modeling for early diagnosis, prognosis, prediction, and monitoring. These competencies extend to the development of clinical decision support tools for managing patients with various pathologies in real-world clinical practice.

GIBI2³⁰ actively participates in clinical trials, exemplified by the establishment of the Imaging Clinical Trials Unit in 2016 (Penadés-Blasco A, et al. Medical imaging clinical trials unit: a professional need. EJR 2022), currently overseeing more than 300 studies.

The group opens its technological resources to facilitate other research groups and individual researchers in achieving their objectives, thereby contributing to the overall scientific quality of diverse projects. These resources are housed within the Experimental Radiology and Imaging Biomarkers Platform (PREBI), expertly managed by the group.

Specialties and expertise of the Node

Population Imaging

GIBI2³⁰ is related to a Datawarehouse at our hospital and research institute, serving as a central access for the efficient storage and retrieval of data, ensuring streamlined research workflows and providing rapid access to critical information.  There is a daily ETL process in place that ensures a continuous update of the data in our systems. Our expertise extends to the integration of data from diverse disciplines such as radiology, pathology, and laboratory. This integrated approach fosters a comprehensive understanding of complex biological processes.

Committed to maintaining high standards, GIBI2³⁰ adheres to rigorous data standardization practices, guaranteeing uniformity in data formats. The Datawarehouse ensures that all the terms are mapped into standards ontologies such as ICD-10, SNOMED or RxNorm, and load the data into the OMOP Common data Model. In addition, this OMOP-CDM is enhanced with the extensions needed for each type of project, being GIBI2³⁰ actively involved in the development of the imaging extension with the OHDSI community. This dedication significantly enhances interoperability and facilitates collaborative research initiatives. Our commitment to standardized, up-to-date data has positioned our data warehouse as a cornerstone for evidence-based decision-making across a spectrum of biomedical research domains.

Thanks to the implementation of this standardized Datawarehouse, GIBI2³⁰ demonstrates the capability to gather data automatically and efficiently. This enables the rapid establishment of retrospective datasets comprising thousands of records, facilitating timely and comprehensive research initiatives within a matter of days.

Artificial Intelligence algorithms and medical imaging

The research group possesses solid knowledge and experience in image processing, as well as the development of Artificial Intelligence algorithms and methodologies for extracting image biomarkers and predicting relevant clinical outcomes. This is its greatest value and contribution to society.

Some of the main methodologies and/or research solutions that the group has established and developed include the following:

• Solutions for the processing of MR, PET, and CT images in oncology and neurodegenerative diseases, including image harmonization, noise reduction, inhomogeneity correction, spatial resampling, intensity normalization, image registration, and lesion segmentation.
• Image analysis algorithms applied to cancer patients, including quantification of ADC, IVIM, Kurtosis, semi-quantitative and pharmacokinetic perfusion in MR; and both radiomics in both MR and CT.
• Tools for studying tumor heterogeneity: Histogram and unsupervised clustering algorithms for the identification and definition of tumor habitats in MR images (Fit-Cluster-Fit).
• Radiogenomic tools for analyzing correlations between image features and genomic data, enabling the discovery of radiomic signatures that can serve as substitutes for genetic tests.
•  Deep Learning algorithms for extracting deep features from MR and/or CT images.
• Image harmonization tools: algorithms based on Generative Adversarial Networks (GAN) for generating synthetic radiological images.
• Convolutional Neural Networks (CNN) and Transformers for the detection and segmentation of organs and tumors from MR and/or CT images.
• Machine Learning models for radiomics and delta-radiomics to predict treatment response in cancer patients.
• Integrative Machine Learning models based on clinical, molecular, and medical image data for predicting overall survival, tumor grade, progression rate, and treatment response.
• End-to-end Deep Learning models for the classification and regression of relevant clinical endpoints.
• Visualization and explainability tools (nosological parametric maps, clustering maps, SHAP values, glyph feature distributions, UMAP projections, etc.) for a better understanding of predictive model results, facilitating their adoption in clinical practice.

Instruments highlights

• 1 compute node (2 x NVIDIA® DGX A100™) and 1 storage node (SERVER SIE LADON GBT CDL 6230 A 2.1 GHz GPU V100), for processing and storing images and biological and clinical data for the development of predictive AI models.
• 1 Multimodal PET/MR (GEHC Signa): simultaneous acquisition of PET and 3T MR with a 60cm bore, allowing rapid acquisition and high spatial resolution. For clinical trials and clinical research projects with patients.
• 1 3T RM (Philips Achieva TX, multi-transmission) with research platforms: 60 cm central tunnel where advanced MR acquisitions are carried out with the possibility of programming special pulse sequences. Both preclinical and clinical
• 1 High resolution multimodality equipment (micro-PET/CT, Bruker Albira) for molecular and anatomical studies in small animals and samples

Additional services offered by the Node

• User-oriented project support with study design and management
• Assistance in image processing, data analyses, and interpretation
• Federated node for distributed analysis
• Data storage
• Computational resources
• Coordination and participation in clinical trials
• 3D printing models
• Interventional radiology training courses in animals

Website: https://www.acim.lafe.san.gva....

Contact: 

carina_soler@iislafe.es 

javier_aquerreta@iislafe.es