Giovanni Gremigna
Uncertainty-Aware and Explainable Deep Learning for Automated Ejection Fraction Estimation and Left Ventricular Segmentation in Video-Based Echocardiography.
Rel. Massimo Salvi, Silvia Seoni. Politecnico di Torino, Corso di laurea magistrale in Ingegneria Biomedica, 2026
|
|
PDF (Tesi_di_laurea)
- Tesi
Accesso limitato a: Solo utenti staff fino al 28 Settembre 2027 (data di embargo). Licenza: Creative Commons Attribution Non-commercial No Derivatives. Download (4MB) |
Abstract
Left ventricular ejection fraction (LVEF) represents a fundamental clinical parameter for the diagnosis, prognostic stratification and management of cardiovascular diseases. Although deep learning models have achieved expert-level performance, their “black-box” nature and the absence of explicit information regarding predictive reliability limit their integration into clinical practice. To address these limitations, this thesis proposes a framework that integrates Uncertainty Quantification (UQ) and Explainable Artificial Intelligence (XAI) techniques into video-based LVEF estimation and left ventricular segmentation, using EchoNet-Dynamic as the methodological baseline. To estimate predictive uncertainty, Monte Carlo Dropout was implemented both in the video-based LVEF regression model and in the spatial segmentation network.
Predictive standard deviation (σ) and normalized entropy along the endocardial border (Border Entropy) were extracted from the two models, respectively
Relatori
Anno Accademico
Tipo di pubblicazione
Numero di pagine
Corso di laurea
Classe di laurea
Aziende collaboratrici
URI
![]() |
Modifica (riservato agli operatori) |
