التفاصيل البيبلوغرافية
| العنوان: |
Federated Learning applied to Brain Tumor Segmentation using MRI Scans ; Aprendizaje federado aplicado a la segmentación de tumores cerebrales mediante resonancias magnéticas ; Aprenentatge federat aplicat a la segmentació de tumors cerebrals mitjançant ressonàncies magnètiques |
| المؤلفون: |
Ysern García, Maria |
| المساهمون: |
Universitat Politècnica de Catalunya. Departament de Teoria del Senyal i Comunicacions, Vilaplana Besler, Verónica, Cumplido Mayoral, Irene |
| بيانات النشر: |
Universitat Politècnica de Catalunya |
| سنة النشر: |
2023 |
| المجموعة: |
Universitat Politècnica de Catalunya, BarcelonaTech: UPCommons - Global access to UPC knowledge |
| مصطلحات موضوعية: |
Àrees temàtiques de la UPC::Enginyeria de la telecomunicació::Processament del senyal::Processament de la imatge i del senyal vídeo, Image processing, Imaging systems in medicine, Electronic data processing--Distributed processing, Computer vision, Neural networks (Computer science), Deep Learning, Federated Learning, Medical image analysis, Brain tumor segmentation, Aprendizaje Profundo, Aprendizaje Federado, Análisis de imagen médica, Segmentación de tumores cerebrales, Imatges--Processament, Imatgeria mèdica, Processament distribuït de dades, Visió per ordinador, Xarxes neuronals (Informàtica) |
| الوصف: |
In recent years, Federated Learning (FL) has emerged as a revolutionary privacy technology that allows training AI models near the data, without moving the data across the boundary of data owners. Initially applied by Google to privacy-preserving prediction of user keystrokes on smartphones, FL was quickly adopted in research of data-driven healthcare and positioned as a key potential solution to the cross-border data utilization challenge. With recent developments in medical imaging facilities, extensive medical imaging data are produced every day. This increasing amoun ; In this thesis we analyze Federated Learning, a novel approach for distributed Deep Learning model training that maintains data privacy. In Federated Learning, a server coordinates the process, and multiple clients possess the data. We conducted a comparison between federated and traditional Deep Learning on a brain tumor segmentation task. We developed a versatile framework that allows for both centralized and federated training, featuring various federated algorithms, with the flexibility to add more. The results indicate that while it is challenging for federated models to reach the performance of centralized ones, it is not impossible. The more local training clients do before the server aggregates their models, the more stable the training becomes, leading to performance improvements. It is worth noting that the drop in performance safeguards patient privacy, while also enabling access to more medical data without violating legal restrictions. ; En esta tesis analizamos el aprendizaje federado, un enfoque novedoso para el entrenamiento distribuido de modelos de aprendizaje profundo que mantiene la privacidad de los datos. En el aprendizaje federado, un servidor coordina el proceso y múltiples clientes poseen los datos. Hemos realizado una comparación entre el aprendizaje federado y el aprendizaje profundo tradicional en una tarea de segmentación de tumores cerebrales. Hemos desarrollado un marco versátil que permite tanto el entrenamiento ... |
| نوع الوثيقة: |
master thesis |
| وصف الملف: |
application/pdf |
| اللغة: |
English |
| Relation: |
http://hdl.handle.net/2117/411580; ETSETB-230.178950 |
| الاتاحة: |
http://hdl.handle.net/2117/411580 |
| Rights: |
S'autoritza la difusió de l'obra mitjançant la llicència Creative Commons o similar 'Reconeixement-NoComercial- SenseObraDerivada' ; Open Access |
| رقم الانضمام: |
edsbas.821DC7C0 |
| قاعدة البيانات: |
BASE |