Author: Erica Tavazzi

Yesterday we had the pleasure to contribute with 3 presentations at the research event “DEI-AI 2024” hosted by our Department of Information Engineering, where we had the opportunity to showcase our latest groundbreaking work.

In our first presentation, titled “Exploring AI Applications in Medicine: Methodological Challenges and Examples,” Barbara Di Camillo, head of Sysbiobig, delved into the intricate world of AI applications in medicine, addressing methodological challenges head-on. From robust feature selection in omics data to dynamic Bayesian networks for simulating disease progression, we showcased cutting-edge methodologies and ongoing projects, including the BRAINTEASER project.

Following that, our second talk, “Artificial Intelligence Methods to Power Real-World Clinical Studies,” was delivered by our researcher Enrico Longato. This presentation underscored the importance of real-world effectiveness in evaluating treatments. We delved into the nuances of retrospective observational studies and discussed innovative AI approaches to combine weighting and matching techniques for robust analysis, a pivotal goal we are actively pursuing within the REDDIE project.

But that’s not all! Our researcher Martina Vettoretti also presented her recently funded project titled “BREATHE – Big data, internet-of-things and aRtificial intelligence to study the impact of personal Exposure to air pollution on AsTHma Exacerbations.” This project, funded under the PRIN 2022 call, aims to revolutionise our understanding of asthma control by studying the impact of personal exposure to air pollution. By leveraging an innovative infrastructure and employing advanced machine learning techniques, Martina and her team are paving the way for the development of prognostic models capable of predicting asthma exacerbations.

These presentations truly underscore our unwavering commitment to advancing AI research in healthcare and driving real-world impact. A heartfelt thank you goes to the event organisers for providing such a platform!

The 7th BRAINTEASER plenary meeting, held at the distinguished Universidade de Lisboa, has just concluded. The event garnered significant participation from various partner institutions, fostering lively discussions and idea exchanges. Attendees, including researchers, clinicians, and communication representatives contributed diverse perspectives, enriching the discourse. The meeting was complemented by an informal social dinner, enhancing networking opportunities.

Our team presented the upgrades and outcomes attained over the past months. We discussed the insights gleaned from examining the effectiveness of integrating pollutant data with clinical variables in Multiple Sclerosis (MS) progression predictive models. Additionally, we detailed the improvements made to both the Amyotrophic Lateral Sclerosis (ALS) and MS progression models. We explored the potential advantages of incorporating embedded stratification into our predictive models, alongside our strategy to ensure the interpretability of artificial intelligence (AI) model results. Lastly, we outlined our plan to exploit Explainable Artificial Intelligence (XAI) to identify and characterise patients for whom providing reliable predictions is challenging.

The collaborative atmosphere facilitated face-to-face meetings and stimulated engaging discussions. As the meeting drew to a close, coordinators and partners expressed optimism about BRAINTEASER’s future role in fostering collaboration and innovation within the global research community. Following two days of intensive exchanges, the consortium is poised to navigate the upcoming months with confidence.

In the context of the H2020 BRAINTEASER project, in collaboration with the University of Pavia and the other consortium partners, we are investigating the possible impact of air pollution on the progression of Multiple Sclerosis (MS).

Starting from the literature, where some potential correlations between pollutant agents and MS emerged, we are analysing the retrospective data provided within the project and made available to the whole research community here.
We explored the use of different machine learning techniques, including both linear and non-linear approaches, combined with the use of manual or automatic techniques for identifying the most robust features for prediction.

From our preliminary analyses on MS, the combination of dynamic environmental features with essential clinical variables appeared to be effective for enhancing the accuracy of predictive models when forecasting the occurrence of a relapse, i.e., an exacerbation of the symptoms. Noticeably, the role of environmental features was confirmed via all used feature selection approaches.

In the context of the H2020 BRAINTEASER project, our group is in charge of developing predictive models for the progression of Amyotrophic Lateral Sclerosis (ALS) and Multiple Sclerosis (MS). In order to identify the most promising approaches to be implemented, we coordinated a systematic review of the artificial intelligence (AI) methodological landscape in ALS, focusing on patient stratification and disease progression prediction, which we performed together with the other project partners.

Out of 1604 reports, we identified 15 studies on patient stratification, 28 on ALS progression prediction, and 6 on both. We highlighted a general agreement in terms of input variable selection for both stratification and prediction of ALS progression, and in terms of prediction targets. A striking lack of validated models emerged, as well as a general difficulty in reproducing many published studies, mainly due to the absence of the corresponding parameter lists. While deep learning seems promising for prediction applications, its superiority with respect to traditional methods has not been established; there is, instead, ample room for its application in the subfield of patient stratification. Finally, an open question remains on the role of new environmental and behavioural variables collected via novel, real-time sensors.

The full article is available here.

These findings laid the groundwork for the development of our models within the project, providing valuable insights into the most effective AI methodologies for patient stratification and disease progression prediction in ALS. They are also guiding our direction in identifying key areas for further development and refinement.