Projects:ARRA:SlicerEM

Back to Slicer ARRA home page

Aim

The EMSegmenter is a state-of-the-art segmentation tool within 3D Slicer. User feedback has reported that clinicians are currently unable to tune the approach to their acquisition protocol, as the user interface is too complex. This proposal addresses this issue by redesigning the user interface, focusing on hiding the complexity of the underlying segmentation algorithm. If successful, this will enable clinicians to automatically segment their own medical scans, even if the corresponding acquisition protocol deviates from the default setting for which the EMSegmenter is optimized.

Research Plan

The EMSegmenter is the result of 15 years of research in medical image segmentation. This has lead to a user interface that exposes a rich set of parameters. These parameters allow the tuning of the EMSegmenter to a wide variety of acquisition sequences. However, tuning these parameters is quite challanging. In addition, Slicer currently does not provide any tools for generating atlases, which are a set of parameters characterizing each structure of interest. We propose to address these issues by creating two user-friendly modules: one for generating the atlases and one for tuning the EMSegmenter to a specific acquisition sequence.

The first module, called Atlas Generator, builds the atlases characterizing each structure of interest. The user simply specifies the training data which can be done via querying XNAT, a database targeted towards medical image analysis. The user also selects the type of information to be extracted from the data. Possible types are the shape, intensity, or relative position of the structures of interest across the training set. Based on this input, the tool automatically generates the atlas.

The second module, called EMSegmenter-Simple, consists of a simple work flow that enables users to adjust the EMSegmenter to their specific acquisition sequence. As part of this proposal, we will create a library of templates, which parametrizes the tool to segmentation tasks frequently ancountered by our user community. User simply adopt the tool to their acquisition scenario by first selecting the proper template. In the second step, the user modifies the value of important parameters of the template. We simplify the tuning of these parameters by providing an instant feedback mechanism. The feedback mechanism updates the automatic segmentation according to the change in the parameter setting. This will allow users to to get an intuitive understanding about impact of certain parameters on the algorithm. In addition, each entry field will be accompanied with a help text. We will also create detailed documentation about the user interface and publish a tutorial for each template.

The project is viewed as successful if a properly trained clinician is able to modify the templates to their acquisition sequence within an hour.

Key Personnel

5% Kilian Pohl

Progress

• 12/11/09 Learned about programming user interfaces in Qt
• 12/04/09 Interviewed applicant for Post Doc position to work on this project
• 11/29/09 Removed bugs related to updated CMake version
• 11/20/09 Organize bugs related to EMSegmenter
• 11/13/09 Tcon featuring the EMSegmenter - participants Andriy Fedorov, Sylvain Jaume, Stuart Wallace, Kilian Pohl - Result of discussion
  • Sylvain will fix EMSEgmenter bugs Slicer 3.5
  • Kilian will fix any EMSegmenter bugs in Slicer 3.4 and earlier
  • Andriy is developing a segmentation pipeline for the Wake Forest Data
  • Stuart will do testing of the EMSegmenter module
• 11/06/09 Meet with Jean-Christophe Fillion-Robin from Kitware to discuss integration of Qt in 3D Slicer
• 10/30/09 Organized Monthly TCon between EMSegmenter developers
• 10/23/09 Organized onsite interview , got in contact with Steve Pieper to discuss next steps, installed Slicer3
• 10/17/09 Started interviewing postdoc as well as solving several HR issues for hiring personal