AI & deep learning for medical image analysis

Speaker: Nicholas Ayache (INRIA)

Abstract: Artificial intelligence and deep learning algorithms have brought about a revolution in medical image analysis, often achieving accuracy comparable to that of medical specialists. It can also help build a digital twin of the patient from his or her medical images and available complementary information (clinical, biological, behavioral, environmental, etc.) This digital twin can be used to guide diagnosis, and then to simulate, optimize and guide therapy. In this talk, I will also illustrate the role of mathematical and biophysical models of anatomy and physiology in forcing these digital medicine algorithms to deliver reliable and interpretable results.

Bio: Nicholas Ayache is a research director at Inria, where he leads the EPIONE research team, dedicated to the digital patient and digital medicine. He is also the Scientific Director of the Interdisciplinary AI Institute 3IA Côte d'Azur, where he holds a research chair. His current research focuses on the introduction of AI algorithms to guide the prevention, diagnosis, prognosis and therapy of patients based on their medical images and all available data.
N. Ayache is a member of the French Academy of Sciences and of the French Academy of Surgery. In 2013-2014 he was a visiting professor at the Collège de France, where he introduced a new course on the "personalized digital patient". He has received numerous prestigious awards including in 2020 the International Steven Hoogendijk Award, in 2014 the Grand Prix Inria-Académie des science, in 2012-2017 an advanced ERC Grant from the European Research Council, in 2008 the Microsoft Grand Prize for Research in Europe (Royal Society), and in 2006 the EADS Foundation Information Science Award.
N. Ayache published over 400 highly cited scientific articles and a dozen of industrial patents, and co-founded seven high-tech companies. He has been a member of several strategic boards in France and abroad.

More about Nicholas Ayache

Deep Learning for Micro-Scale Imaging: Challenges and Advances in Osteoporosis Analysis

Speaker: Eleonora D'Arnese (School of Informatics, University of Edinburgh)

Abstract: Although medical imaging is a fundamental tool for studying the genesis and the evolution of multiple pathologies, it generates a considerable amount of data that proves to be burdensome to analyze. Deep learning has emerged as a valuable support tool for clinical practice focusing on repetitive tasks like segmentation. This talk will focus on the study of osteoporosis by analyzing images at the micro-scale - where the pathology originates - and, in particular, on the classification and segmentation of such data. It will also focus on micro-scale images' challenges and possible directions for applying Deep Learning to such data.

Bio: Eleonora D’Arnese received her Ph.D. in Information Technology from Politecnico di Milano in 2023. Additionally, she got her B.Sc. and M.Sc. in Biomedical Engineering from Politecnico di Milano in 2016 and 2018, respectively and in 2018 she also received her M.Sc. in BioEngineering from the University of Illinois at Chicago. She worked as a Postdoctoral Researcher at Politecnico di Milano, and at Università Vita-Salute San Raffaele. She is currently a Lecturer in Biomedical AI at the School of Informatics, University of Edinburgh and her research interests mainly revolve around medical image analysis and computer vision.

More about Eleonora D’Arnese

Deep Learning in Medical Imaging for Radiotherapy: Applications and Challenges

Speaker: Daniele Loiacono (Politecnico di Milano)

Abstract: Deep learning is quickly transforming the field of medical imaging and has shown great potential in radiotherapy. This talk provides an overview of how deep learning techniques might enhance several tasks in radiotherapy — including image segmentation, dose distribution prediction, and treatment planning — to improve accuracy and efficiency in treatment procedures. The presentation will also address the main challenges faced in this domain, such as data annotation, model interpretability, and integration into clinical workflows.

Bio: Daniele Loiacono was born in Lecco, Italy, in 1980. He graduated cum laude in 2004 in Computer Engineering and he received a Ph.D. in Computer Engineering in 2008 from Politecnico di Milano. He is currently an associate professor at the Department of Electronics and Information of Politecnico di Milano. His main research interests include machine learning, evolutionary computation, games and AI applications for healthcare.

More about Daniele Loiacono

On how to let Osteocytes Teach SR-MicroCT Bone Lacunae Segmentation

Tutors: Eleonora D'Arnese (University of Edinburgh), Isabella Poles (Politecnico di Milano)

Abstract: Synchrotron Radiation micro-Computed Tomography (SR-microCT) is a promising imaging technique for osteocyte-lacunar bone pathophysiology study. However, acquiring them costs more than histopathology, thus requiring multi-modal approaches to enrich limited/costly data with complementary information. This tutorial will present LOTUS, a novel histopathology-enhanced disease-aware distillation model for bone microstructure segmentation from SR-microCTs. Exploring its components, we will discuss its peculiarity and applicability to different datasets while providing an overview of single-, multi-modal, and state-of-the-art distillation methods for image segmentation.

Bio: Isabella Poles received her B.Sc. and M.Sc. degrees in Biomedical Engineering from Politecnico di Milano in 2019 and 2022, respectively. In 2022, she also got her M.Sc. in Bioengineering at the University of Illinois at Chicago. In 2024, Isabella was a Research Trainee at the Mahmood Lab at Harvard Medical School and Brigham and Women's Hospital in Boston. Currently, she is a third-year Ph.D. student in Information Technology at NECSTLab of Politecnico di Milano. Her research focuses on developing a deep learning-based framework for image analysis in biopsy, integrating multi-scale and multi-modal images to support clinicians and researchers in their pathology analysis needs.

Implementing 3D neural networks on advanced neuroimaging approaches to monitor disease progression and outcome after traumatic brain injury

Tutors: Marcello De Salvo, Federico Moro, Elisa R. Zanier (Mario Negri Institute)

Abstract: In this tutorial, we will provide an overview of traumatic brain injury (TBI), exploring the role of rodent models in TBI research as valuable tools for studying the mechanisms of secondary injury evolution commonly observed in patients. Among various neuroimaging techniques, magnetic resonance imaging (MRI) stands out for its strong translational potential, providing detailed insights into both primary injuries and the progression of secondary damage. We will introduce challenges in conducting preclinical MRI studies with a focus on strategies to perform brain segmentation to monitor the evolution of brain damage in selected vulnerable regions. After the introduction, we will conduct a hands-on session where participants will explore the structure and implementation of a 3D multi-task CNN for segmenting TBI rodent brains.

Bio: Marcello De Salvo is a researcher at the Unit of Pathobiology and Neuroimaging in the Laboratory of Traumatic Brain Injury and Neuroprotection @ Mario Negri Institute for Pharmacological Research IRCCS
Federico Moro is the head of the Unit of Pathobiology and Neuroimaging in the Laboratory of Traumatic Brain Injury and Neuroprotection @ Mario Negri Institute for Pharmacological Research IRCCS
Elisa R. Zanier is the head of the Department of Acute Brain and Cardiovascular Injury and the head of the Laboratory of Traumatic Brain Injury and Neuroprotection @ Mario Negri Institute for Pharmacological Research IRCCS

Denoising microscopy images with deep-learning

Tutors: Melisande Croft, Diya Srivastava, Igor Zubarev (Human Technopole)

Abstract: Denoising is a useful image processing step that can drastically improve downstream analysis. Because microscopy images are often subjected to noise from the imaging process and from the detectors themselves, a wide range of methods have been developed to reduce its impact on post-processing and quantification. Deep-learning algorithms have established themselves as the most performant approaches in that regard and are nowadays in common use in scientific image processing. In this tutorial, we will focus on Noise2Void, a widely used deep-learning algorithm that does not require ground-truth images to train. Using its implementation in the CAREamics library, we will discuss its use in the context of scientific analysis and its limitations, as well as describe the current landscape of existing algorithms.

Bio: Melisande Croft is Research Software Engineer @ National Facility for Data Handling and Analysis, Human Technopole
Diya Srivastava is Data Scientist and Research Software Engineer @ Jug Group, Human Technopole
Igor Zubarev is Bioimage Analyst and Research Software Engineer @ Jug Group, Human Technopole

Segmenting microscopy images with shallow- and deep-learning

Tutors: Eugenia Cammarota, Damian Dalle Nogare (Human Technopole)

Abstract: Image segmentation is a fundamental step in many image analysis workflows. The separation of images into foreground and background classes (semantic segmentation), or into individual objects of a given class (instance segmentation) is critical to enable downstream data processing and quantification by separating an image into computationally tractable regions or objects. In recent years, both classical (shallow) machine learning approaches and deep learning approaches have been leveraged to perform this task. In this tutorial, we will consider two methods for image segmentation. Firstly, we will perform pixel classification (semantic segmentation) using a random forest approach implemented in the Labkit plugin for Fiji/ImageJ. Secondly, will use the cellpose package, which leverages deep learning to generate an instance segmentation of microscopy data. We will discuss the application and limitations of these technologies in the context of scientific image analysis.

Bio: Eugenia Cammarota is Bioimage Analyst and Research Software Engineer @ Bioimage Analysis Infrastructure Unit, National Facility for Data Handling and Analysis, Human Technopole
Damian Dalle Nogare is Manager Image Analysis Facility @ Bioimage Analysis Infrastructure Unit, National Facility for Data Handling and Analysis, Human Technopole

Beyond Labels: Advancing Medical Imaging Through Unsupervised Learning for Anomaly Detection, Vessel Segmentation, and Phenotype Discovery

Tutor: Soumick Chatterjee (Human Technopole)

Abstract: Unsupervised learning and limited annotation methods have emerged as powerful paradigms in medical imaging, enabling automated feature extraction and pattern recognition while reducing reliance on extensive manual annotations. This tutorial talk will explore state-of-the-art techniques for three key applications: anomaly detection, vessel segmentation, and phenotype extraction. Unsupervised learning methods, such as autoencoders and generative adversarial networks, can model the distribution of normal anatomical structures, facilitating the detection of pathological deviations without the need for labelled abnormal samples. For vessel segmentation, both unsupervised and weakly supervised approaches have shown promise, with self-supervised learning and clustering-based techniques leveraging intrinsic image patterns, while weakly supervised methods utilise limited annotations, such as scribbles or image-level labels, to enhance segmentation accuracy. Finally, representation learning techniques, including autoencoders and contrastive learning, enable the extraction of clinically relevant phenotypic features from medical images, supporting disease subtyping, biomarker identification and Genotype-phenotype associations. This session will provide a comprehensive discussion of these methods, highlighting their role in advancing medical imaging analysis. A hands-on tutorial will accompany the talk, demonstrating practical implementations and evaluation strategies to equip researchers with the necessary tools to apply unsupervised and weakly supervised learning effectively in medical imaging.

Bio: Soumick Chatterjee is a post-doc researcher @ Glastonbury Group, Genomics Research Centre, Human Technopole, and a Lecturer in AI for Medical Imaging @ Faculty of Computer Science, Otto von Guericke University Magdeburg, Germany

REALITY-CENTRIC AI

Speaker: Mihaela van der Schaar (University of Cambridge)

Abstract: We define Reality-centric AI as AI which can operate effectively, reliably, and accountably in the real world. We believe that the balance between the two camps needs to change. Therefore, we propose a reality-centric research agenda consisting of eight pillars to pull together disparate fields of current research and build the necessary new ML tools and models to deliver the world-changing promise of AI.

Bio: Mihaela van der Schaar is the John Humphrey Plummer Professor of Machine Learning, Artificial Intelligence and Medicine at the University of Cambridge. In addition to leading the van der Schaar Lab, Mihaela is founder and director of the Cambridge Centre for AI in Medicine (CCAIM). Mihaela was elected IEEE Fellow in 2009 and Fellow of the Royal Society in 2024. She has received numerous awards, including the Johann Anton Merck Award (2024), the Oon Prize on Preventative Medicine from the University of Cambridge (2018), a National Science Foundation CAREER Award (2004), 3 IBM Faculty Awards, the IBM Exploratory Stream Analytics Innovation Award, the Philips Make a Difference Award and several best paper awards, including the IEEE Darlington Award. She was a Turing Fellow at The Alan Turing Institute in London between 2016 and 2024. Mihaela is personally credited as inventor on 35 USA patents, many of which are still frequently cited and adopted in standards. She has made over 45 contributions to international standards for which she received 3 ISO Awards. In 2019, a Nesta report determined that Mihaela was the most-cited female AI researcher in the U.K.

More about Mihaela van der Schaar

EVIDENCE BASED TREATMENT OF CARDIOVASCULAR DISEASE: THE ROLE OF REAL-WORLD DATA AND MACHINE LEARNING

Speaker: Alicia Uijl (Amsterdam UMC)

Abstract: We will discuss the use of electronic health records in real-world evidence with several use cases in cardiovascular disease.

Bio: Alicia is an epidemiologist specialised in Cardiovascular Epidemiology. Currently, she is an Assistant Professor at Amsterdam University Medical Center and affiliated researcher at Karolinska Institutet, Sweden. In her research, Alicia focuses on applying advanced epidemiological methodology and data science on real-world data sources to obtain real-world evidence for patients with cardiovascular diseases, with a particular interest in heart failure.

More about Alicia Uijl

Assessing brain diseases using smartphones

Speaker: Oliver Y. Chen (Lausanne University Hospital, University of Lausanne)

Abstract: About one in six people worldwide suffer from neurological disorders. Traditional brain disease assessments are typically performed by a physician during infrequent clinic visits, often involving expensive and time-consuming data acquisition techniques, such as magnetic resonance imaging (MRI). In contrast, smartphone-based assessments enable remote, frequent evaluation at home without supervision and with minimal patient burden. The relatively low cost and increasing availability of smartphone devices, compared to the cost and waiting times for hospital equipment and shortage of medical professionals, make digital health solutions accessible and affordable to a large population. Here, I present two examples of how smartphones can be used to assess Parkinson’s disease and multiple sclerosis.

Bio: Oliver Yibing Chén has a multidisciplinary academic background in engineering (PhD, Oxford), neurobiology (Honorary Fellow, University College London), psychology (Research Associate, Yale), biostatistics (Master, Johns Hopkins), theoretical statistics (Research Assistant, Northwestern) and mathematical statistics (Master, Washington). He has worked not only in industry, at Axio Research in Seattle (statistics) and at F. Hoffmann-La Roche in Basel (Pharma Research and Early Development - pRED), but also in the clinic, at the Universitair Ziekenhuis Brussel (neurology) and the Nationaal MS Centrum (multiple sclerosis).
From 2021 to 2022, he was Assistant Professor in Artificial Intelligence and Big Data Analytics at the University of Bristol, before being granted tenure in September 2022. He then joined Switzerland in early 2023 where he is Chief of the Platform of Bioinformatics at Lausanne University Hospital (CHUV), and Assistant Professor on conditional pre-tenure at the level of Associate Professor in Artificial Intelligence and Statistical Science at University of Lausanne (UNIL).

More about Oliver Y. Chen

Mastering Real-World Health Data in Regional Healthcare Databases: Challenges and Opportunities

Tutors: Katherine Logan, Laura Savaré (Human Technopole)

Abstract: This tutorial will focus on electronic health records held in regional health databases in Italy, with a particular emphasis on the crucial role of real-world data (RWD) in generating empirical evidence essential for informed decisions in clinical and policy contexts, especially within the Italian National Health System (NHS). We will cover the structure of these datasets, strategies for overcoming common challenges, and offer practical, worked examples from a variety of fields, such as pharmacoepidemiology. Additionally, there will be applied examples in the cardiovascular field, demonstrating how RWD can drive insights into cardiovascular health and treatment outcomes. A key objective is to explore advanced statistical methodologies that fully harness RWD, providing a more accurate reflection of real-world clinical practice and addressing the limitations of controlled clinical studies, which often struggle to capture the diverse behavioral dynamics of patients and the real-world effectiveness of treatments. The tutorial will highlight the unique challenges in this field and present the most suitable data analysis techniques for maximizing the value of RWD in medical research and policy.

Bio: Katherine Logan is a PhD student in Data Analytics and Decision Sciences @ Di Angelantonio & Ieva Group, Health Data Science Centre, Human Technopole & Politecnico di Milano
Laura Savaré is a post-doc researcher @ Di Angelantonio & Ieva Group, Health Data Science Centre, Human Technopole

Smart Remote Monitoring

Tutors: Luigi della Torre, Alessandro Gumiero (STMicroelectronics)

Abstract: The importance of remote monitoring is dramatically growing following the evolution of the technologies and the request for a medicine closer to the user. The COVID period highlighted the necessity to have something simple capable to acquire vital signs everywhere, continuously and correctly. The amount of data collected by this kind of devices allows to develop AI models augmenting the reliability of the parameters computed and precision of the feedbacks. Key topics:

• Remote monitoring
• Edge AI
• Wearable sensors

An example of what it means to design and create a medical device for remote monitoring is presented, keeping particular attention to user needs and medical grade of the signals and parameters calculated.

Bio: Luigi della Torre is System R&D Health Remote Monitoring & Sensors Application Manager at STMicroelectronics, involved in the development of health remote monitoring low power E2E platforms and system algorithm for biometrics signal acquisition.
Alessandro Gumiero is electronic designer at STMicroelectronics, involved in the development of innovative integrated electron systems in the medical and fitness field.

Revolutionizing Maternal and Child Care: The Power of Real-World Evidence

Tutor: Anna Cantarutti (Laboratory of Healthcare Research and Pharmacoepidemiology, University of Milano-Bicocca)

Abstract: Real-World Evidence (RWE) has emerged as a powerful tool in healthcare research, offering valuable insights into the real-world effectiveness and safety of interventions. This review explores the potential of RWE to revolutionize maternal and child care. By leveraging data from electronic health records, clinical registries, and other real-world sources, researchers can better understand disease burden, treatment patterns, and patient outcomes. This knowledge can inform evidence-based decision-making, optimize clinical practice, and improve patient outcomes. In this workshop, we will delve into the methodological challenges associated with observational studies, particularly immortal time bias and confounding. We will also explore practical applications of integrating diverse Real-World Data (RWD) sources, including the implementation of artificial intelligence in clinical practice and the use of geospatial analysis to assess area-level socioeconomic disparities.

Bio: Anna Cantarutti is Assistant Professor (RTDB) in Medical Statistics @ Laboratory of Healthcare Research and Pharmacoepidemiology, University of Milano-Bicocca

Personalized care pathways and optimization of the emergency room resources thanks to the analysis of health records

Tutors: Luca Barbanotti (SAS), Jacopo Lenkowicz, Carlotta Masciocchi (Fondazione Policlinico Universitario Agostino Gemelli IRCCS)

Abstract: In this tutorial we will show how it is possible to respond to the demand for services generated in an emergency department, in a way that is effective for the health of patients and efficient in the operational management of a hospital, through collaboration with services and departments involved in the emergency: the emergency/urgency sector to try to predict the impact of the influx of cases in the short term; the inpatient and intensive care units to analyze the diagnostic/therapeutic information collected from the electronic medical records; the management control for reporting and optimal resource allocation; and research, to prepare the tools for computational analysis with AI algorithms.
By exploiting the information obtained from the SAS analytics, optimal resource planning in the short and medium term is enabled and answers are given, responding with customised paths for each patient.

Bio: Luca Barbanotti is Senior Customer Advisor in Advanced Analytics and AI. He's leading SAS AI advisory initiatives across multiple industries and since 2024 He's part of SAS EMEA & AP Healthcare Industry Board Leadership Team.
Jacopo Lenkowicz is a physicist with a PhD in Oncological Sciences, specializing in data science and machine learning for healthcare. Experienced in working with medical imaging and integrating data from data warehouses to support clinical decision-making and research. Focused on improving healthcare outcomes through data-driven approaches.
Carlotta Masciocchi is a biomedical Engineer with a PhD in Oncological Sciences. She has been involved in national and international projects regarding Radiomics and Machine Learning applications in Radiotherapy and Oncology and she is currently the coordinator of the Gemelli Generator Real World Data laboratory at Fondazione Policlinico Gemelli. She focuses her research activity on the development of privacy-preserving algorithms, healthcare terminological systems and development of predictive models using data from Electronic health records.

Data re-use for public good: a collaboration between Fondazione Cariplo and Novartis Farma

Tutors: Diana Pozzoli (Fondazione Cariplo), Naomi Viapiana (Novartis), Saverio D'Amico (Humanitas University), Bernardo Nipoti (University of Milano-Bicocca)

Abstract: Data are an essential part of our daily lives. When high quality data is sourced, made accessible and reused in a responsible way, it can help us all make more informed decisions. In this tutorial we will explore a concrete case of data re-use for public good starting with the fruitful collaboration between Fondazione Cariplo and Novartis Pharma S.p.A. The two entities launched a call for proposals to fund projects investigating the representativeness gap between the populations enrolled in clinical trials and the general population affected by the same pathology. Starting from the dataset provided by Novartis Pharma S.p.A. applicants have been required to formulate original working hypotheses and to develop models and guidelines to foster more effective treatments for all. Specific projects funded under the call will be illustrated during the presentation.

Bio: Diana Pozzoli is Deputy Director of the Science and Technology area of Fondazione Cariplo
Naomi Viapiana is Biostatistician at Novartis
Saverio D'Amico is Data Scientist and Machine Learning Engineer at Humanitas University
Bernardo Nipoti is Associate Professor of Statistics at Department of Economics, Management and Statistics - University of Milano-Bicocca