An integrated platform for high-quality data-driven research

Abstract

Abstract The digital transformation of healthcare has led to an unprecedented growth in the volume and heterogeneity of clinical data. At the same time, aging populations and rising demands on healthcare systems require cost-effective, scalable solutions. Artificial intelligence (AI) has emerged as a powerful tool in clinical research, enabling the extraction of complex diagnostic features from data that are otherwise imperceptible. However, the development and deployment of AI models demand structured, large-scale, high-quality datasets. The extensive preparation required before data can be used for AI training (collection, transfer, cleaning, translation, standardization, etc.) poses a significant bottleneck between data acquisition and algorithm development. This work presents the development of a platform designed to accelerate data-driven clinical research, with a special focus on the creation of AI models. The platform was built with a modular architecture, enabling flexibility and scalability across different research domains. It includes a data pipeline that supports the ingestion, validation, anonymization, and secure storage of various types of clinical data —ranging from medical images (e.g. US, MRI), biosignals (e.g. ECG), and structured tables (e.g. sociodemographic or laboratory data). It supports interoperability through the adoption of domain standards (file formats, ontologies), and facilitates collaboration through role-based access and intuitive web-based interfaces. The system supports both on-premise and cloud deployments and has been conceived to ensure compliance with data protection regulations by design. Our platform is currently in active use in 18 ongoing research projects —most of which involve the development and validation of AI algorithms— by over 60 users, including clinicians and researchers across three continents. Clinicians can easily explore the data, perform quality control and generate ground-truth annotations through intuitive web-based interfaces, while researchers benefit from standardized data inputs and can deploy, test, and refine their models. The complete platform infrastructure has been successfully deployed in two major cloud services as well as on-premise in two hospitals (with further deployments in progress) and has already processed more than 400,000 clinical data samples, demonstrating its robustness and scalability in real-world research environments. In conclusion, the proposed integrated environment streamlines the complex workflow of multimodal clinical studies and supports agile, collaborative development of AI tools. By aligning technical infrastructure with clinical needs and regulatory requirements, the platform significantly reduces barriers in translational AI research and promotes faster, more reliable innovation in healthcare.

Ähnliche Arbeiten