A major innovation is taking shape within our rehabilitation hospital: a digital platform designed to integrate clinical data, artificial intelligence, scientific literature and, in the future, digital twins into neurorehabilitation pathways. The project has been developed to support the work of the entire rehabilitation team and to foster an increasingly shared approach to clinical management, facilitating the planning, monitoring and evaluation of integrated rehabilitation processes
Thanks to the Fit4AI-SR project, developed within Fit4MedRob (Fit For Medical Robotics), neurorehabilitation at Villa Beretta is becoming increasingly data-driven. This does not mean that data will replace clinicians; rather, it will help them better understand and interpret the complexity of each patient.
By integrating clinical data, laboratory results, diagnostic imaging and an extensive digital library, the project is set to enhance the daily work of healthcare professionals across the organization. Most importantly, it aims to improve the personalization and effectiveness of treatments for patients with neurological conditions, particularly those recovering from stroke, traumatic injuries, and neurodegenerative diseases.
We spoke with Stefano Sangalli, engineer at the SCAI Consortium and lead developer of the platform, and Eleonora Guanziroli, engineer and Head of the Bioengineering Department at our research institute.
FROM AUTOMATED DATA COLLECTION TO THE CLINICAL DATA LAKE
The first step is the automated collection of clinical data. To make this possible, a platform has been created to retrieve information from a wide range of sources: electronic health records, hospital admissions, PDFs, robotic devices and sensors, motion analysis systems, diagnostic imaging, blood tests, X-rays, and clinical or textual data produced by healthcare professionals.
The value lies not only in the quantity of data collected, but in the ability to bring them together within a single system.
This process is enabled by a digital agent, a sort of intelligent collector that synchronizes with the admissions database, identifies who is currently hospitalized or discharged, and retrieves the available content. In this way, information no longer remains scattered across different archives, documents, and systems, but enters a continuous acquisition flow.
At the core of the platform is the structured clinical data lake, designed to manage and correlate heterogeneous information in a secure and interoperable way. It is not simply a file repository: documents are segmented, cleaned, normalized, and transformed into information that can be used by artificial intelligence.
This is a decisive step, because an AI model cannot work directly on thousands of unordered pages. If the input data are inaccurate, incomplete, or poorly structured, the final output also risks being unreliable.
A PLATFORM FOR THE REHABILITATION TEAM
The platform will become part of the daily work of the entire rehabilitation team. Physicians, physiotherapists, bioengineers, neuropsychologists, psychologists, and nurses will have access to a highly innovative way of interacting in the management of complex care pathways, helping to personalize them as effectively as possible.
They will be able to query the system in natural language and obtain clinical summaries, tables, graphs, assessment scale data, and links to scientific literature. The system, in fact, does not work only on patient data, but also on a digital library composed of articles, PDFs, and scientific content that has already been processed. This makes it possible to connect the individual clinical history of a patient with the knowledge available in the scientific literature.
The artificial intelligence models are designed with Explainable AI systems, so that the recommendations generated are as transparent and interpretable as possible for clinicians. This aspect is essential in a healthcare context, where obtaining an answer is not enough: it must also be possible to understand the elements on which that answer is based.
Another important step is the creation of patient clusters. The platform can group patients with similar characteristics, for example based on assessment scales, diagnosis, treatments received, and outcomes achieved. This makes it possible to observe recovery trajectories, compare different profiles, and generate more targeted suggestions.
TRACKING VILLA BERETTA’S TECHNOLOGICAL EVOLUTION
The implementation of Fit4AI-SR also makes it possible to revisit the clinical and technological history of Villa Beretta through three main phases: a first phase, from 2008 to 2014, with a more limited contribution from technology; a second phase, from 2014 to 2019, marked by the introduction of a greater number of technological tools; and a third phase, from 2019 to 2026, characterized by further growth in both the quantity and complexity of the technologies available.
This perspective allows us to observe not only how the use of devices has changed over time, but also how patient assessment has evolved, with new clinical scales, greater integration among different professional figures, and a more articulated measurement of the rehabilitation pathway.
Recovering this historical data opens the way to retrospective studies that are far richer than traditional models. The aim is not simply to compare patients before and after the introduction of a technology, but to understand how complex, multimodal rehabilitation programs have influenced the multidimensional assessment of each patient.
A MORE PRECISE PATIENT ASSESSMENT
One of the most relevant aspects concerns the possibility of improving the precision of clinical assessment. In many studies, different patients are considered comparable because their lesions are located in apparently similar areas. However, greater integration of diagnostic imaging and a more accurate quantification of lesion location could reveal differences that are currently underestimated.
This means that two patients considered similar may in fact have very different profiles and require different rehabilitation strategies. This is where the platform can influence clinical reasoning: not by imposing a decision, but by offering a deeper interpretation of the available data.
HOW THE WORK OF ALL OF US WILL CHANGE
In the daily work of all members of the rehabilitation team, the platform may become a tool to support shared decision-making in treatment selection. Even before reaching the digital twin stage, the system will make it possible to retrieve clinical data, recovery trajectories, assessment scales, previous treatments, and related scientific literature. This may help physicians formulate more informed hypotheses and assess more precisely which pathway to propose to the patient.
The platform also includes operational tools for daily use, such as interactive clinical dashboards, real-time progress monitoring, automated reports, and therapeutic performance indicators. These tools can help professionals follow the patient’s evolution and adapt treatment based on the data collected.
The system may also interact with robotic devices and support the optimization of the parameters used to deliver therapeutic exercise. From this perspective, AI could help identify settings that are better suited to each individual patient.
Nursing work could also be facilitated by infrastructural developments that are apparently simple but highly concrete. One example is the use of Bluetooth systems to automatically transfer daily recorded parameters, such as blood pressure and other clinical data, into the medical record. This would reduce the time spent on manual transcription and lower the risk of error, freeing up resources for activities with greater care value.
THE DIGITAL TWIN AS A FUTURE PERSPECTIVE
One of the most innovative developments of the project concerns the clinical digital twin, a dynamic digital replica of the patient that will be able to virtually reproduce their functional status and the evolution of their rehabilitation pathway.
Based on the data collected, the system will be able to simulate different scenarios, such as a change in pharmacological therapy, robot settings, or the overall rehabilitation approach. In the future, professionals will be able to apply a program or therapeutic modification to the patient’s digital twin, observe the simulation of the outcome, and then decide whether to transfer that choice into real clinical practice.
The digital twin will thus become a virtual testing space, useful for evaluating different scenarios without directly exposing the patient to unnecessary attempts. The final decision will always remain with the clinician, but the system will be able to provide a broader information base to build increasingly personalized and predictive rehabilitation pathways.
SECURITY, VALIDATION AND CLINICAL RESPONSIBILITY
From a technological perspective, the project is based on a hybrid-cloud and modular architecture, with local processing systems at Villa Beretta and computational components hosted in dedicated data centers. This configuration ensures healthcare data security, platform scalability, and the possibility of integrating new devices and technologies over time.
The platform will grow progressively: the more data are collected, the more the system can refine its analyses. However, this growth also requires a validation process. Professionals will need to assess the quality of the responses, indicate whether information is useful or not, and thus contribute to progressively improving the system.
Clinical responsibility remains central. Artificial intelligence does not make decisions in place of the physician: it highlights data, correlations, and possibilities that professionals can integrate into their own reasoning. The therapeutic choice always remains human.
