Artificial Intelligence in Hemodialysis: from failures detection to patient-therapy analysis

In recent decades, technological innovation has accelerated rapidly, leading to the spread of advanced devices and technologies. This trend has also had a profound impact on the biomedical sector, where such tools are increasingly being used to optimize clinical activities, improve prevention, diagnosis, and treatment, and reduce healthcare costs. As a result, research has progressively shifted toward the development of innovative, technologically advanced, yet economically sustainable investigative solutions. Artificial Intelligence (AI) has become a valuable tool for the analysis of biomedical data and for supporting the development of advanced medical technologies. The main advantage of AI methods, when applied to physiological signals or clinical information, lies in their ability to automatically learn meaningful patterns and correlations from large datasets without requiring extensive manual processing or controlled laboratory experiments. This brings considerable benefits in terms of time, costs, and reduction of the operational burden associated with traditional analysis workflows. Additionally, AI techniques enable the precise and quantitative characterization of biomedical data, allowing for an exploration of how changes in model architecture, training strategies, or input features impact predictive accuracy and reliability. In recent years, equipment manufacturers have shown a growing interest in directing their scientific efforts toward topics that can help address practical problems in the field. In the present Ph.D. project, the use of artificial intelligence is proposed as a tool to improve hemodialysis therapies, both from the dialysis machine side and the patient side. Hemodialysis is a blood-purification treatment performed outside the body and prescribed to patients with advanced kidney dysfunction. The therapy relies on dialyzers, which are devices containing bundles of hollow synthetic fibers. These fibers act as semi-permeable barriers that facilitate the transfer of unwanted solutes and metabolic by-products, such as urea and creatinine and many others, from the bloodstream, thereby helping to restore biochemical balance. With the purpose to fill some major gaps in the literature, the specific activities performed in this thesis propose the use of artificial intelligence to develop smarter dialysis systems, particularly those related to enhancing prediction, early identification, and overall management of clinical conditions. In this context, AI approaches hold considerable potential in nephrology, as they can enhance diagnostic accuracy, support therapeutic decision-making, and contribute to a more effective and personalized management of renal replacement therapies.