Project

Back to overview

Longitudinal Brain Tumor Segmentation with Uncertainty Estimation using Fully-connected Conditional Random Field and Perturb-and-Maximum-Posterior-Marginal Estimation

English title Longitudinal Brain Tumor Segmentation with Uncertainty Estimation using Fully-connected Conditional Random Field and Perturb-and-Maximum-Posterior-Marginal Estimation
Applicant Reyes Mauricio
Number 169607
Funding scheme Project funding (Div. I-III)
Research institution Institute for Surgical Technology and Biomechanics Universität Bern
Institution of higher education University of Berne - BE
Main discipline Other disciplines of Engineering Sciences
Start/End 01.04.2017 - 31.03.2020
Approved amount 177'963.00
Show all

All Disciplines (3)

Discipline
Other disciplines of Engineering Sciences
Biomedical Engineering
Clinical Cancer Research

Keywords (5)

Glioma; Longitudinal brain image analysis; MRI; Medical image analysis; Brain cancer

Lay Summary (German)

Lead
Dieses Projekt befasst sich mit der automatischen Identifikation verschiedener Gewebetypen in Bildern der Magnetresonanztomographie von Gehirntumorpatienten. Diese Identifikation, auch Segmentierung genannt, ermöglicht die Quantifizierung der Volumina und weiterer Gewebsparameter, welche für die Diagnose wie auch Planung und Verfolgung der Therapie von grosser Bedeutung sind. Eine automatische Segmentierung der Tumorgewebe erfolgt durch rechnergestützte Methoden, die gegenüber manueller Segmentierung viel effizienter und auch objektiv sind. Die aktuell verfügbaren Methoden weisen einen Mangel an räumlicher Konsistenz ihrer Segmentierungsergebnisse und ein Fehlen einer Schätzung der Messunsicherheit derselben Ergebnisse auf.
Lay summary

Inhalt und Ziel des Forschungsprojekts

Das Ziel dieses Projekts ist es eine Methode zu entwickeln, welche die Messunsicherheit unserer bisherigen Gehirntumorsegmentierungsmethode schätzen kann. Diese Methode wird auf einem Fully-connected Conditional Random Field (CRF) und Perturbations-basiertem Sampling beruhen. Das Fully-connected CRF wird eine bessere räumliche Konsistenz der Segmentierungsergebnisse garantieren, während ein neuartiger perturbations-basierter Ansatz das gleichzeitige Schätzen der Messunsicherheit erlauben wird.

Wissenschaftlicher und gesellschaftlicher Kontext des Forschungsprojekts

Das Schätzen der Messunsicherheit in der automatischen Tumorsegmentierung wird es in Zukunft erleichtern manuelle Korrekturen der Segmentierungsergebnisse vorzunehmen oder unsichere Messergebnisse bei einer weiteren für den Patienten relevanten Analyse auszuschliessen. Die Korrektur von Tumorsegmentierungen ist von grosser Bedeutung in der Radiotherapie und Neurochirurgie, deren Planung auf einer exakten räumlichen Delineation des Tumors basieren.

Direct link to Lay Summary Last update: 06.01.2017

Responsible applicant and co-applicants

Employees

Name Institute

Project partner

Publications

Publication
Analyzing the Quality and Challenges of Uncertainty Estimations for Brain Tumor Segmentation
Jungo Alain, Balsiger Fabian, Reyes Mauricio (2020), Analyzing the Quality and Challenges of Uncertainty Estimations for Brain Tumor Segmentation, in Frontiers in Neuroscience, 14, 282.
Assessing Reliability and Challenges of Uncertainty Estimations for Medical Image Segmentation
Jungo Alain, Reyes Mauricio (2019), Assessing Reliability and Challenges of Uncertainty Estimations for Medical Image Segmentation, in Medical Image Computing and Computer Assisted Intervention – MICCAI 201922nd International Conferenc, Shenzhen Springer International Publishing, China.
On the Effect of Inter-observer Variability for a Reliable Estimation of Uncertainty of Medical Image Segmentation
Jungo Alain, Meier Raphael, Ermis Ekin, Blatti-Moreno Marcela, Herrmann Evelyn, Wiest Roland, Reyes Mauricio (2018), On the Effect of Inter-observer Variability for a Reliable Estimation of Uncertainty of Medical Image Segmentation, in Medical Image Computing and Computer Assisted Intervention – MICCAI 201821st International Conferenc, Springer International Publishing, Spain.
Towards Uncertainty-Assisted Brain Tumor Segmentation and Survival Prediction
Jungo Alain, McKinley Richard, Meier Raphael, Knecht Urspeter, Vera Luis, Pérez-Beteta Julián, Molina-García David, Pérez-García Víctor M., Wiest Roland, Reyes Mauricio (2018), Towards Uncertainty-Assisted Brain Tumor Segmentation and Survival Prediction, in Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries, Granada, Spain474-485, Springer International Publishing, Granada, Spain474-485.
Uncertainty-driven Sanity Check: Application to Postoperative Brain Tumor Cavity Segmentation
Jungo Alain, Meier Raphael, Ermis Ekin, Herrmann Evelyn, Reyes Mauricio (2018), Uncertainty-driven Sanity Check: Application to Postoperative Brain Tumor Cavity Segmentation, in 1st Conference on Medical Imaging with Deep Learning (MIDL 2018), Amsterdam In Press, MIDL OpenReview, Amsterdam In Press.

Collaboration

Group / person Country
Types of collaboration
Technical University of Munich - Prof. Menze Germany (Europe)
- in-depth/constructive exchanges on approaches, methods or results
- Publication
Olea Medical Technologies France (Europe)
- in-depth/constructive exchanges on approaches, methods or results
- Industry/business/other use-inspired collaboration
Dana Farber Cancer Institute - Prof. Aerts United States of America (North America)
- in-depth/constructive exchanges on approaches, methods or results
- Publication
- Exchange of personnel
Swiss Glioma Network Switzerland (Europe)
- in-depth/constructive exchanges on approaches, methods or results
- Publication
- Research Infrastructure
MD Anderson Cancer Center - Dr. Rao United States of America (North America)
- in-depth/constructive exchanges on approaches, methods or results

Awards

Title Year
Second position in the 2017 International Brain Tumor Segmentation Challenge: Survival Prediction 2018

Associated projects

Number Title Start Funding scheme
180365 Predict and Monitor Epilepsy After a First Seizure: The Swiss-First Study 01.02.2019 Sinergia

Abstract

Background:Automated brain tumor segmentation from multimodal MR images has the potential to leverage diagnosis, treatment, patient follow-up, radiomics studies, etc. However, despite progress in brain tumor image segmentation current algorithms still lack spatial consistency and do not provide a measure of certainty of their results, which is critical for adequate exploitation of results in subsequent steps.Methods:In the proposed project we aim at developing an automated multimodal MR brain tumor segmentation. Based on preliminary results, we will extend our previously proposed supervised-based segmentation approach with a fully-connected Conditional Random Field (CRF) regularization module, featuring improved segmentation accuracy and spatial consistency of tumor compartments. To produce measures of uncertainty of the segmentation results, we will further develop a perturb-and-maximum-posterior-marginal estimation technique. Both improvements feature low computational costs, in agreement with the requirements of the clinical workflow. The approach is to be evaluated on both local glioma datasets and publicly available databases used for benchmarking of brain tumor segmentation algorithms. The evaluation will lead us to test the following hypotheses: Hypotheses:H0) Employing a Fully Connected Conditional Random Field (CRF) approach for brain lesion segmentation yields improved segmentation results, in terms of accuracy and spatial consistency, as compared to a standard pair-wise CRF.H1) Segmentation uncertainty can be obtained through a perturbation & MPM approach under a fully-connected CRF setup. Expected added value of the proposed study:The project will provide improved accuracy and spatial consistency for automated quantification and delineation of brain tumor compartments. The ability of the proposed approach to yield uncertainty measures of the segmentation opens up opportunities to create time-effective human-machine interfaces necessary to monitor and correct results of the automated approach. This can have a substantial impact in clinical scenarios such as radio-therapy and neuro-surgery, where a precise delineation of the tumor is needed in order to precisely target the tumor while preserving as much healthy tissues as possible. In turn, uncertainty estimates of the segmentation can be used as data-quality measures for high-throughput radiomics studies relying on these segmentation results to extract imaging biomarkers of diagnosis, assessment of prognosis, and prediction of therapy response.
-