Design and implementation of MLOps-based healthcare pipelines using the Cybersecurity Framework

Recent advances in Artificial Intelligence (AI) are accelerating the adoption of Machine Learning (ML) solutions in the healthcare sector, where compliance with stringent regulations (e.g., GDPR and HIPAA) poses significant challenges throughout the development lifecycle. Thanks to its broad set of practices and tools, Machine Learning Operations (MLOps) emerges as a promising approach to address these issues. As a result, MLOps-based pipelines can play a pivotal role in managing ML models effectively, which is crucial for supporting diagnostic and prognostic activities. However, healthcare systems must also address cybersecurity requirements, which are often unexplored or only partially considered in ML lifecycle management. In this work, we investigate the design and implementation of an MLOps-based healthcare pipeline developed for the DARE (DigitAl lifelong pRevEntion) foundation. The proposed environment incorporates security mechanisms aligned with the NIST Cybersecurity Framework (CSF) 2.0. For this reason, we first describe the pipeline architecture and the mechanisms adopted to address cybersecurity requirements. Then, using a systematic mapping-based approach, we analyze the feasibility of our pipeline in ensuring several CSF functions, with particular emphasis on data security, detection, and recovery. In detail, we show that adopting MLOps principles enables the pipeline to directly support Data Security, while other CSF categories related to Detect and Recover can be addressed indirectly and may require additional implementation steps, including dedicated tools or external providers. Finally, through dedicated use cases, we show how requirements indirectly ensured can be concretely implemented within the proposed pipeline.