Chiara Marzi

Fixed-term Assistant Professor working in a different University

Alma Mater Studiorum - Università di Bologna

Research

Keywords: fractal dimension, MRI, CT, structural complexity, artificial intelligence, machine learning

Feature Selection Stability in High-Dimensional Data Analysis: Applications in Biomedical Research: this thesis focuses on feature selection stability techniques specifically tailored for high-dimensional data analysis. The study will involve evaluating the stability and performance of various feature selection methods when applied to biomedical data, especially MRI-derived features, aiming to identify robust biomarkers or signatures associated with specific diseases or conditions.

Explainability of Machine Learning Models from a Statistical Perspective: Enhancing Interpretability and Trust in Predictive Models: this thesis aims to explore the concept of explainability in machine learning models, focusing on statistical methods for improving the interpretability and transparency of these models. The study will investigate various techniques, such as feature importance analysis, model-agnostic interpretability methods, and post-hoc explainability approaches, to provide insights into the underlying factors that contribute to the predictions made by machine learning algorithms. The research will involve evaluating the effectiveness and reliability of these explainability methods in different domains, with a particular emphasis on biomedical applications. By developing a deeper understanding of the statistical aspects of model explainability, this thesis seeks to enhance the trustworthiness and acceptance of machine learning models in real-world decision-making processes.

Causal Machine Learning for Biomedical Imaging: Exploring the Role of Causal Inference in Biomedical Applications: this thesis aims to investigate the application of causal machine learning techniques in the field of biomedical imaging. The study will focus on leveraging causal inference methods to analyze complex biomedical data, particularly in the context of medical imaging modalities such as MRI and CT. The research will involve developing and applying causal machine learning algorithms to identify causal relationships between imaging features and clinical outcomes, enabling a deeper understanding of the underlying mechanisms and disease progression.

Bayesian Methods for Computing Fractal Dimension in Neuroimaging: Advancing the Measurement of Tissue Structural Complexity: this thesis aims to further develop and refine Bayesian methods for computing fractal dimension in the field of neuroimaging. Building upon a pilot study, the research will focus on enhancing the accuracy and precision of measuring tissue structural complexity through advanced Bayesian approaches. The study will explore the use of Bayesian modeling techniques, such as Markov Chain Monte Carlo (MCMC) methods, to estimate the fractal dimension of neuroimaging data, particularly in the context of brain MRI. The research will involve refining the computational algorithms, optimizing parameter selection, and evaluating the performance of the Bayesian method on larger datasets. The ultimate goal is to improve our understanding of tissue complexity and provide more reliable and robust measurements of fractal dimension in neuroimaging studies. This thesis will contribute to advancing our knowledge of neuroimaging analysis techniques and their applications in characterizing brain structure. In collaboration with Professor Diciotti.