11th Nov - Multiple Sclerosis


Understanding to collaborate. 
An multidisciplinary approach to the disease progression: a collaboration example about Multiple Sclerosis






Francesca Grassi, Department of Physiology and Pharmacology "Vittorio Erspamer" 

Laura Palagi, Department of Computer, Control and Management Engineering “Antonio Ruberti”     


Multiple sclerosis is the leading cause of progressive neurological disability in young people, as it mainly affects people between the ages of 20 and 50, with very high human and social costs. The disease begins with a relapsing-remitting phase (RR), which evolves into a secondarily progressive form (SP) over an extremely variable period, which remains largely unpredictable. This is increasingly frustrating, as nowadays several therapies that can prevent relapses even for a long time are available, although the possible adverse effects are the more serious the more effective the drug is. An early prediction of disease course would make it possible to differentiate treatment based on the expected aggressiveness of the disease, reserving high-impact therapies to patients at greater risk.

To increase prognostic capacity, approaches based on Machine Learning (ML) algorithms have been attempted, but levels of specificity and sensitivity adequate for clinical use have not yet been attained. However, most of these studies are based on highly specialized data, not used in normal clinical practice, so that the machines would hardly become a widely used tool, even after marked improvements in their the predictive performance.

In this seminar we present different aspects of a collaboration aimed at exploring the possibility of using data available in common clinical practice to predict medium term disease course. The team puts together physicists and engineers expert in Machine Learning and decision support systems, neurologists and neurophysiologists from four different departments of Sapienza University. We will describe the initial efforts required to create a transversal knowledge base in all the participants to the study and define a common language that would allow effective communication. We will then illustrate how we worked on a database containing the clinical records of the patients attending the Sant'Andrea hospital in Rome, to make it usable for machine learnerns. Finally, we will present the results obtained using a predictive model that uses two different learning paradigms, one based on the clinical information related to a single visit (Visit-oriented), the other using the sequence of visits available for the patient (History-oriented) and possible ways of combining the two. We will also briefly describe the ML techniques used and the performance indicators used for evaluating the reliability of the model. Finally we will discuss the results obtained, which show how "real" data can be effectively used to predict the evolution of MS and suggest possible strategies to improve the results of the analysis, through an active collaboration between those who collect the data and those who analyse them.



mercoledì 11 novembre 2020, ore 16.00 - 18.00


Per partecipare occorre collegarsi tramite Google Meet al link meet.google.com/poy-szhq-ate


© Università degli Studi di Roma "La Sapienza" - Piazzale Aldo Moro 5, 00185 Roma