Italian Researchers Propose National AI Network for Obesity Monitoring

A team of Italian clinicians and artificial intelligence researchers has published a detailed conceptual proposal for a national digital infrastructure that would connect obesity treatment centers across Italy into a single, AI-supported surveillance ecosystem. The paper, appearing in Eating and Weight Disorders – Studies on Anorexia, Bulimia and Obesity, introduces OBESI-NET, short for Osservatorio Basato su Ecosistema Sanitario Digitale di Intelligenza Artificiale per l’Obesità – NETwork Nazionale. Crucially, the authors are explicit that nothing has been built yet: the work is a Level V expert opinion and conceptual project proposal, not a report of an implemented system, deployed software, or measured clinical results.

The motivation rests on a data fragmentation problem. According to the proposal, clinical, anthropometric, behavioural, technological and epidemiological information about obesity in Italy is currently scattered across registries, observational studies, electronic health records and local institutional initiatives. In their view, this fragmentation limits longitudinal monitoring, national comparability, and the development of data-driven tools for prevention, diagnosis and clinical management. OBESI-NET is proposed as a software ecosystem that would link clinical centers of the Italian Society of Obesity (SIO), healthcare professionals and AI engineering groups, supporting standardized collection, secure integration and analysis of heterogeneous data ranging from clinical and behavioural records to imaging, wearable sensor readings and, in future developments, multi-omics data.

The epidemiological backdrop the authors cite is substantial. Drawing on the latest ISTAT survey from 2025, they report that 46.4 percent of the Italian population over 18 is overweight. While that figure appears relatively stable, the same report indicates obesity has risen over the last decade from 9.8 percent to 11.6 percent. Men are more affected than women, with 55.1 percent of men and 38.2 percent of women affected by overweight or obesity. The authors frame this increase as clinically relevant because obesity is a chronic disease associated with multiple comorbidities, complications and healthcare costs, and they note that obesity prevalence globally has more than doubled since 1990.

The technical core of the proposal is multimodal artificial intelligence applied to longitudinal real-world data. The authors argue that obesity-related data are heterogeneous, high-dimensional and longitudinal, which makes machine learning and multimodal AI a plausible methodological response: such tools could support integration of different data sources, identification of patient subgroups, prediction of risk trajectories and development of decision-support tools. They are careful, however, to state that AI should not be considered a stand-alone solution. Its clinical usefulness, they write, depends on data quality, representativeness, external validation, interpretability, bias assessment, clinician oversight and integration into existing care pathways.

Privacy protection occupies a central place in the proposed architecture. The plan calls for federated learning, a distributed training approach in which clinical and sensitive data remain locally stored at the two reference centers in Milan and Palermo while AI algorithms are transferred, trained and updated in a distributed manner. The entire process would be supervised by SIO centers overseeing the procedures adopted for individual tasks. According to the authors, this approach would allow the development of robust and generalizable predictive models without centralizing data, reducing privacy risks and supporting compliance with the General Data Protection Regulation. They cite numerous studies indicating that federated learning is an effective solution for collaborative analysis of healthcare data while maintaining security and statistical reliability.

The two proposed reference sites reflect existing Italian expertise. The Palermo unit would build on the Human-Machine Interaction Laboratory, which has developed a framework called IODeep that extends the DICOM Information Object Definition to accommodate the weights of a deep neural network, enabling automated analysis of medical images and biomedical data directly within hospital PACS servers and supporting federated learning by centralizing the model rather than the data. The group there also works on drug discovery, drug repurposing and retrieval-augmented generation systems for clinical decision support. The Milan unit would be hosted at the Humanitas AI Center at IRCCS Humanitas Research Hospital, a multidisciplinary center of physicians, data scientists, engineers and bioinformaticians working on machine learning, omics data analysis, computer vision and natural language processing for healthcare.

The Milan ecosystem extends into the European research landscape. Train, a 2023 spin-off of the Humanitas AI Center, develops generative AI models for synthetic data generation, patient digital twins and clinical activity optimization. Both the center and Train participate in SYNTHIA, a project funded under the Innovative Health Initiative of Horizon Europe that brings together more than thirty partners to develop federated platforms for generating realistic, multimodal synthetic clinical, genomic, imaging and digital device datasets while preserving patient privacy, with validation across domains including cancers, neurodegenerative diseases and type 2 diabetes.

The proposal positions itself against existing international projects. The European BigO project built an interactive platform collecting heterogeneous data on about 3,700 children from 33 schools and 2 clinics across five countries. NoHoW tested a digital toolkit with wearable devices for weight-loss maintenance as an interventional trial. BIO-STREAMS, ongoing, aims to create a multinational pediatric obesity database with AI risk prediction. GroHealth W8Buddy delivers specialist weight management services across multiple NHS centers. Outside Europe, the International Weight Control Registry functions as a research registry rather than a clinically embedded platform, the China Diabetes Registry focuses on metabolic disease broadly rather than obesity-centered AI, and the Hong Kong Diabesity Register targets combined diabetes and obesity management with AI limited to a chatbot. None, the authors argue, constitutes a national obesity-centered surveillance system embedded in routine clinical practice.

The authors also devote attention to organizational and human factors, describing OBESI-NET as a socio-technical infrastructure in which digital tools, clinical responsibilities, organizational processes and patient engagement develop together. Drawing on recent Italian experiences with integrated care for frail older adults, organizational health literacy and professional training, they argue that implementation would require shared protocols, multidisciplinary coordination, continuous training covering data quality, privacy protection and AI literacy, and feedback mechanisms from clinicians and patients. A national physician network of nutritionists, endocrinologists, general practitioners and specialists would collect information through desktop and mobile digital tools, transforming routine clinical activity into structured data.

The paper’s conclusions are candid about its own status. Because OBESI-NET has not been implemented, the work reports no technical deployment, prospective patient recruitment, clinical outcomes, validation studies or measured effects on healthcare processes. Future work would need to address feasibility, interoperability with existing electronic health records, data governance, sustainability, clinician engagement, patient recruitment and multicenter validation, and any AI model developed within the framework would require external validation, bias assessment, transparency and clinical interpretability before use in decision support. If implemented and validated, the authors write, OBESI-NET could provide a framework for epidemiological monitoring, clinical research, treatment stratification and public health planning in obesity care.

Subject of Research: A conceptual proposal for an Italian national AI-supported obesity monitoring and precision medicine network

Article Title: OBESI-NET: an Italian network proposal for AI-supported obesity monitoring and precision medicine

Article References: Gambino, O., Savesvki, V., Manco, M., D’Amico, S., Ranzini, M. B. M., Contino, S., Pirrone, R., Buscemi, S., & Vettor, R. (2026). OBESI-NET: an Italian network proposal for AI-supported obesity monitoring and precision medicine. Eating and Weight Disorders – Studies on Anorexia, Bulimia and Obesity, 31(1), Article 92. https://doi.org/10.1007/s40519-026-01907-8

Image Credits: AI Generated

DOI: 10.1007/s40519-026-01907-8

Keywords: obesity, artificial intelligence, federated learning, precision medicine, digital health, Italy, epidemiological surveillance, real-world data, SIO, multimodal data, GDPR, health informatics