Computer vision in medicine and beyond: deep learning approaches to solve real-world problems

Carlini, Gianluca (2025) Computer vision in medicine and beyond: deep learning approaches to solve real-world problems, [Dissertation thesis], Alma Mater Studiorum Università di Bologna. Dottorato di ricerca in Fisica, 37 Ciclo.
Documenti full-text disponibili:
[thumbnail of main.pdf] Documento PDF (English) - Accesso riservato fino a 1 Settembre 2025 - Richiede un lettore di PDF come Xpdf o Adobe Acrobat Reader
Disponibile con Licenza: Creative Commons: Attribuzione - Non Commerciale - Non Opere Derivate 4.0 (CC BY-NC-ND 4.0) .
Download (170MB) | Contatta l'autore

Abstract

Deep Learning has transformed the way we analyze and interpret complex data, enabling the discovery of intricate patterns and the automation of specialized, time-consuming tasks. This thesis explores its application across various domains, with a primary focus on medical imaging but also extending to earth sciences. By leveraging computer vision, this work demonstrates how deep learning can extract meaningful information from images to address challenging problems effectively. The research primarily employs Convolutional Neural Networks (CNNs) combined with advanced image processing techniques to enhance data quality and interpretability. Several case studies illustrate the versatility of these methods. In the medical field, deep learning is applied to electron microscopy images of kidney biopsies, where it enables precise segmentation and measurement of anatomical structures, achieving results comparable to expert assessments. Another application focuses on analyzing Whole Slide Images in the context of myeloid disorders, providing valuable prognostic insights through large-scale image analysis. The thesis also addresses the segmentation of pulmonary airways in lung fibrosis, a particularly complex task for both human experts and AI models. The proposed approach, developed as part of an international challenge, ranks among the top-performing solutions. Beyond the medical domain, deep learning is used to interpret sedimentary core images, demonstrating its ability to automate expert-driven analysis and contributing to the creation of a publicly available dataset in this field. The methods developed in this thesis achieve state-of-the-art performance, often matching or surpassing human expertise, and underline the importance of interdisciplinary collaboration in advancing AI-driven solutions. Future directions include refining the proposed models, extending their applicability to new challenges, and integrating transformer-based architectures to enhance the processing of multimodal data.

Abstract
Tipologia del documento
Tesi di dottorato
Autore
Carlini, Gianluca
Supervisore
Dottorato di ricerca
Ciclo
37
Coordinatore
Settore disciplinare
Settore concorsuale
Parole chiave
Computer Vision, Deep Learning, Neural Networks, Medicine
Data di discussione
21 Marzo 2025
URI

Altri metadati

Gestione del documento: Visualizza la tesi

^