Project

Back to overview

MRF-based Fusion Framework for the Automated Parcellation of Human Cerebral Cortex

English title MRF-based Fusion Framework for the Automated Parcellation of Human Cerebral Cortex
Applicant Thiran Jean-Philippe
Number 144334
Funding scheme Project funding (Div. I-III)
Research institution Laboratoire de traitement des signaux 5 EPFL - STI - IEL - LTS5
Institution of higher education EPF Lausanne - EPFL
Main discipline Information Technology
Start/End 01.10.2012 - 31.03.2013
Approved amount 29'255.00
Show all

Keywords (9)

Non-Rigid Registration; Cerebral Cortex; Head and Neck; Radiotherapy; Medical Image Processing; Markov Random Fields (MRF); Lymph Nodes; Parcellation; Atlas-based Segmentation

Lay Summary (English)

Lead
Lay summary

This is a six-month extension to the previous three-year SNSF proposal (Grant 205321-124797).

Atlas-based segmentation is a key area of research that has a significant impact on diverse medical imaging applications. It has been shown in many recent works that automated segmentations obtained by the fusion of results from multiple atlases are more accurate and reliable than the results obtained from a single atlas. In our recent works, we developed a Markov Random Field (MRF) based fusion framework that generalizes many of the existing fusion methods; we have also proposed new atlas-fusion strategies and evaluated various fusion methods in the context of segmentation of lymph nodes in the Head and Neck (H&N) CT images. 

The objective of the current project is to extend our MRF-based framework from a regular 3-D grid to a more general graph structure; this allows expanding our framework to other important applications, in particular, to the parcellation human cerebral cortex. We will be performing a comprehensive evaluation of various fusion methods for the parcellation of cerebral cortex. 

More precisely we want to propose an extension of our current MRF-based fusion framework to a more general graph structure. Such a nice theoretical framework has several potential clinical applications; in particular, we want to focus on the application of “parcellation of human cerebral cortex”. We will be performing a comprehensive evaluation of various fusion methods that also incorporate an additional edge-preserving MRF-based smoothness term. Further, we would like to use this framework for the connectivity analysis of the baby brains as well. Depending on the results from different fusion methods, this work could ultimately result in constructing a more accurate general template for the parcellation of the human cerebral cortex.


Direct link to Lay Summary Last update: 21.02.2013

Responsible applicant and co-applicants

Employees

Publications

Publication
Multi-­‐atlas fusion methods for segmentation of head and neck lymph nodes for radiotherapy planning
Gorthi S., Bach Cuadra M., Villafruela Vicario J., Tercier P.-A., Thiran J.-Ph., Allal A. (2013), Multi-­‐atlas fusion methods for segmentation of head and neck lymph nodes for radiotherapy planning, in Proceedings of the 2nd ESTRO Forum, GenevaESTRO, Geneva.

Collaboration

Group / person Country
Types of collaboration
Hôpital Fribourgeois Switzerland (Europe)
- in-depth/constructive exchanges on approaches, methods or results

Awards

Title Year
PhD thesis of Mr S. Gorthi

Associated projects

Number Title Start Funding scheme
124797 Adaptive atlas-based segmentation of head and neck lymph nodes in CT images for accurate radiotherapy planning 01.04.2009 Project funding (Div. I-III)

Abstract

This is a six-month extension proposal to the previous three-year SNSF proposal (Grant 205321-124797).Atlas-based segmentation is a key area of research that has a significant impact on diverse medical imaging applications. It has been shown in many recent works that automated segmentations obtained by the fusion of results from multiple atlases are more accurate and reliable than the results obtained from a single atlas. In our recent works, we developed a Markov Random Field (MRF) based fusion framework that generalizes many of the existing fusion methods; we have also proposed new atlas-fusion strategies and evaluated various fusion methods in the context of segmentation of lymph nodes in the Head and Neck (H&N) CT images. The objective of the current proposal is to extend our MRF-based framework from a regular 3-D grid to a more general graph structure; this allows expanding our framework to other important applications, in particular, to the parcellation human cerebral cortex. We will be performing a comprehensive evaluation of various fusion methods for the parcellation of cerebral cortex. In the rest of this section, we first briefly summarize the contributions made during our previous project, which are relevant to the current proposal, and then, present the goals of our current proposal.Contributions during previous project:Intensity modulated radiotherapy (IMRT) is the high precision technique to accurately deliver X-ray radiation treatment for different tumor locations. The major bottleneck for its implementation in the routine clinical practice is the inability to accurately and automatically delineate the lymph node regions. Atlas-based segmentation methods are well suited for the segmentation of such structures with very low contrast with respect to the surrounding structures. In 2009, we developed an adaptive atlas selection strategy for the segmentation of H&N CT images (mandible and brainstem) and proposed to use both pixel-based forces and region-based forces from selected structures for a more accurate segmentation. From the clinical point view, in order to transfer the automated segmentation results to clinical software, we also developed a DICOM exporter. A comparison of single atlas selection versus some of the popular atlas-fusion methods for the segmentation of soft tissue structures in the Head & Neck region (left parotid & right parotid glands) is done in 2010. A new variational framework for performing atlas-based segmentation is proposed and evaluated in 2011. A generalized MRF-based framework for performing fusion of segmentation results obtained from multiple atlases is proposed. In another relevant work, various recent energy minimization methods have been evaluated for solving MRF-based model. Recently, a comprehensive evaluation of various atlas fusion strategies for the H&N lymph nodes segmentation is performed.Objectives of the current proposal:We want to propose an extension of our current MRF-based fusion framework to a more general graph structure. Such a nice theoretical framework has several potential clinical applications; in particular, we want to focus on the application of “parcellation of human cerebral cortex”. We will be performing a comprehensive evaluation of various fusion methods that also incorporate an additional edge-preserving MRF-based smoothness term. Further, we would like to use this framework for the connectivity analysis of the baby brains as well. Depending on the results from different fusion methods, this work could ultimately result in constructing a more accurate general template for the parcellation of the human cerebral cortex.
-