FROM CLOUD TO BEDSIDE

C. Tacchetti^1,2 | ¹Medical School & S.RACE (San-Raffaele Artificial Intelligence Center);Vita-Salute University San Raffaele, Milan, Italy; ²Experimental Imaging Center, IRCCS Hospital San Raffaele, Milan, Italy

Our ability to treat diseases has long been based on the concept of treating the specific pathology. Therapeutic protocols have traditionally been defined using a purely statistical approach, measuring treatment effectiveness on large populations of patients ostensibly suffering from the same type of disease. However, the efficacy of a therapeutic protocol is rarely 100%, and varying percentages of patients suffering from a specific disease do not respond to ‘standard of care’ treatment, despite apparently carrying the same pathology as those who do respond. The recognition of the relevance of individual variability in prognostic prediction and therapeutic response, coupled with the awareness that disease and patient are inseparable, forms the basis of so-called precision or personalized medicine. The search for descriptive parameters of this uniqueness, i.e., parameters able to best predict the characteristics of a given disease in a given patient, has consequently led to an impetuous increase in data, largely acquired or stored in digital format. The use of this enormous amount of data for predictive purposes (prognostic and/or therapeutic) increasingly requires sophisticated analysis techniques, and artificial intelligence (AI) can be the answer to this need. Leveraging a partnership with Microsoft, the San-Raffaele AI Center of Excellence (S-RACE) has engineered and implemented a secure, trustworthy, and responsible by-design interoperability platform on Azure cloud.¹ This platform, built on the FHIR protocol (the international standard for healthcare data processing), is designed to revolutionize how healthcare data is managed and analyzed. The platform’s capabilities include retrieving, classifying (based on major medical ontologies), and analyzing multidimensional realworld data encompassing clinical reports, laboratory results, pathology, imaging, and omics. Crucially, integrated AutoML systems will enable the automatic generation of powerful black-box models for research, while parallel development of glass (white)box models will ensure transparency and explainability in decision making. This comprehensive process is fully traceable, reproducible, and strictly compliant with all major AI regulations (GDPR, ISO 42001, AI-Act, RMF). Currently 21 different projects are running on the platform to develop predictive models in the oncology, neurology, cardiovascular, diabetes areas.