Towards reliable brain tumor segmentation in MRI neuroimaging: integrating uncertainty estimation and ensemble methods for clinical applications

This thesis addresses uncertainty in automated brain tumor segmentation using deep learning. An ensemble of SwinUNETR, SegResNet, and Attention U-Net models was developed, incorporating test-time augmentation and test-time dropout for uncertainty estimation. Evaluation on the BraTS 2021 dataset show...

Descripción completa

Detalles Bibliográficos
Autor: Mosinska, Agata
Tipo de recurso: tesis de maestría
Fecha de publicación:2025
País:España
Institución:Universitat Politècnica de Catalunya (UPC)
Repositorio:UPCommons. Portal del coneixement obert de la UPC
Idioma:inglés
OAI Identifier:oai:upcommons.upc.edu:2117/430325
Acceso en línea:https://hdl.handle.net/2117/430325
Access Level:acceso abierto
Palabra clave:Deep learning (Machine learning)
Imaging systems in medicine
Image segmentation
Brain--Tumors
Segmentació automatitzada de tumors cerebrals
Ensembles d'aprenentatge profund
Estimació de la incertesa
Augmentació en temps d'inferència
Dropout en temps d'inferència
SwinUNETR
SegResNet
Attention U-Net
Predicció conscient del risc
BraTS 2021
Segmentació d'imatges mèdiques
Automated brain tumor segmentation
Deep learning ensembles
Uncertainty estimation
Test-time augmentation
Test-time dropout
Risk-aware prediction
Medical image segmentation
Aprenentatge profund
Imatgeria mèdica
Imatges--Segmentació
Cervell--Tumors
Àrees temàtiques de la UPC::Informàtica::Intel·ligència artificial::Aprenentatge automàtic
Descripción
Sumario:This thesis addresses uncertainty in automated brain tumor segmentation using deep learning. An ensemble of SwinUNETR, SegResNet, and Attention U-Net models was developed, incorporating test-time augmentation and test-time dropout for uncertainty estimation. Evaluation on the BraTS 2021 dataset showed that the ensemble achieved statistically significant improvements in segmentation accuracy, particularly for edema, relative to individual models. Uncertainty maps correlated with voxel-wise error and enabled risk-aware prediction. A Streamlit application was prototyped for clinical feasibility, allowing for interactive visualization of segmentation and uncertainty. The developed application lays the theoretical and technical groundwork for future compliance studies in medical image segmentation, with the potential to accelerate regulatory approval and clinical integration. Further progress will require larger, more diverse models and prospective reader studies to translate these preliminary findings into clinically meaningful improvements.