At the end of the course, students will have acquired advanced knowledge, skills, and competencies related to the theoretical and practical foundations of AI applied to medicine. Students will: • Have a deep understanding of the main sources of biomedical data, pre-processing and transformation techniques, performance and model evaluation methods, and emerging challenges in the application of AI in clinical settings. • Explore heterogeneous clinical datasets, identifying typical problems related to data quality and structure; apply techniques for cleaning, transforming and representing biomedical data; implement efficient and effective machine learning and deep learning pipelines, validate and test models rigorously, comprehensively evaluating classification/regression performance and assessing bias and generalization. • Use development tools (e.g., Python, scikit-learn, TensorFlow) to design and document robust and generalizable algorithmic solutions; collaborate on hands-on projects, contributing to the design and presentation of real-world data-driven solutions. • Be able to integrate computer science, statistics and biomedical knowledge to contribute to the development of intelligent systems for early diagnosis, continuous monitoring and personalized medicine; work in interdisciplinary settings, communicating effectively with health care and technology professionals; evaluate the ethical, legal and social impacts of developed solutions.

Python programming; notions of statistics, machine learning, deep learning

• Module 1: Biomedical Data Sources (9 hours) The main sources of medical data are introduced and carefully described, including structured data (i.e., electronic health records), physical/physiological signals (electrocardiography, electroencephalography, electromyography, photoplethysmography, electrodermal activity, inertial signals), and imaging (radiography, computed tomography, magnetic resonance). Particular attention is devoted to acquisition systems, data format, resolution, dimensions, units of measurement, noise sources, sampling rate, frequency range, typical signal shape, and clinical relevance. An overview of the most relevant public datasets is provided. • Module 2: Pre-processing (9 hours) Data cleaning methods are presented, including filtering, missing values and outliers detection and imputation. Pre-processing approaches are described, from normalization to data transformation, from feature extraction to feature selection. • Module 3: Artificial intelligence (9 hours) AI algorithms are presented, from simple (fuzzy systems) to complex (deep neural networks). Several machine learning models are described, with a particular focus on their hypothesis on data structure, potential and limitations, and practical tips. An overview of the training and optimization settings for complex deep learning algorithms is provided, and practical suggestions are given for developing effective, robust, and bias-free prediction models. • Module 4: Performance and model evaluation (9 hours) Performance evaluation metrics for classification and regression problems are described, along with their pros and cons. A comprehensive overview of validation and test methods is given, along with important considerations on their applicability and efficacy. Issues on generalization ability are discussed, as well as possible solutions. • Module 5: Challenges and Opportunities (9 hours) Limitations of supervised learning strategies are discussed, including handling data and label scarcity. Privacy-preserving methods (e.g. federated learning) are introduced, essential for medical applications. Attention will be given to interpretability and explainability of health-oriented diagnostic/monitoring systems. Bias identification and mitigation strategies are described. Technical requirements for edge processing and real-time prediction are listed. The potential integration of generative AI in clinical decision support systems is discussed. • Laboratories: 12 hours Laboratories will cover all theoretical concepts and methods, applying these in real-world medical problems. These include data pre-processing, machine/deep learning model training and optimization, performance evaluation, and fairness, explainability, and generalizability. • Seminars: 3 hours Seminars will provide an in-depth overview of AI applications on different data sources for specific objectives. These include AI-based analysis of multi-modal medical data for better understanding, early diagnosis, and monitoring of acute and chronic disorders.

• Classroom lectures (45 hours): theoretical lectures dedicated to the introduction and in-depth study of the fundamental concepts of AI applied to the medical context. • Exercises and practical labs (12 hours): computer lab activities aimed at the practical application of the theoretical content covered. • Thematic seminars (3 hours): meetings with researchers and clinical staff, dedicated to the presentation of real case studies and applications of AI in the diagnostic, therapeutic or monitoring fields. • Individual or group project: students will develop an application project, individually or in small groups, aimed at solving a real problem using the techniques learnt. The project will be assessed in the final examination.

• Lecture slides: made available weekly on the online teaching platform, to support classroom explanation. • Interactive notebooks (Python/Jupyter): used during exercises and laboratories to explore machine learning and deep learning algorithms on clinical data. • Handouts and summary sheets: containing theoretical summaries, glossaries of terms and conceptual diagrams. • Recorded video lectures: available for each course module, accessible asynchronously on the e-learning platform. • Multimedia material and links to open-access resources: scientific articles, public medical datasets, online courses, and visual content for self-study. Supplementary material includes: [1] Aurélien Géron, Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow, 2nd ed., O'Reilly Media, 2019 (introduce and explore the main techniques of machine learning and deep learning with practical examples) [2] E. Topol, Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again, Basic Books, 2019 (provides application and critical context on the impact of AI in clinical practice)

Lecture slides; Lecture notes; Lab exercises; Video lectures (current year); Multimedia materials;

Exam: Written test; Individual project; Group project;

The examination consists of two parts: 1. Written examination (duration: 1 hour, maximum score: 20): Multiple-choice test consisting of approximately 20 questions covering the theoretical and application topics covered during the course. It is not permitted to consult texts, notes or electronic devices connected to the Internet. 2. Individual or group application project (duration 20 minutes, maximum score: 10): Development of a project with a real medical dataset, in which machine learning/deep learning methods covered in the lecture are applied and evaluated. Honours are awarded to students who achieve maximum marks in both components and demonstrate excellence, originality and advanced mastery of the topics. This will be assessed though specific questions during the project presentation. The examination modalities are kept the same in all appeals, even if the candidate refuses a previous mark.

In addition to the message sent by the online system, students with disabilities or Specific Learning Disorders (SLD) are invited to directly inform the professor in charge of the course about the special arrangements for the exam that have been agreed with the Special Needs Unit. The professor has to be informed at least one week before the beginning of the examination session in order to provide students with the most suitable arrangements for each specific type of exam.