Permalink
Switch branches/tags
Find file Copy path
Fetching contributors…
Cannot retrieve contributors at this time
65 lines (41 sloc) 3.3 KB

Back to Projects List

Using Extensions

Key Investigators

  • Derek Gillies (Robarts Research Institute)

Project Description

Supervised machine learning methods require an input image with a corresponding mask for training a network. Currently, the Volume Clip extension is used with the Scene Views module to create these masks, but it is a multi-step process. This project will focus on automating these steps to create masks for applicators in 2D ultrasound liver images.

Objective

  1. Create a module that outputs masks for machine learning applications

Approach and Plan

  1. Dedicate time to learning
  2. Talking to more knowledgable people

Progress and Next Steps

  • 3D images were loaded in as a Vector Volume, 2 points were used as inputs for tip and trajectory computation, then a needle model was created

  • The Volume Clip to Model module was used on the needle model to change the image to be white inside and black outside

  • Since 2D images were loaded into Slicer as a Vector Volume, the Vector to Scalar Volume module was used to convert the images to use with 3D logic

  • 2D images do not have pixel spacing encoded, so a few "ctk.ctkDoubleSpinBox()" entry fields were added to accept user input

  • TODO: The Volume Clip to Model does not always fill the inside of the model, but the cause is unknown since it is an inconsistent issue

  • TODO: Deleting a node with slicer.mrmlScene.RemoveNode() does not restart the naming convention (i.e. creating the variable again will still add "Model_1" instead of just "Model")

  • TODO: The needle model extends outside the ultrasound filed of view. The mask should be cropped to be present only where there is image data

Illustrations

Inputs to the module are an ultrasound image with a point on the shaft and a point on the tip: Inputs to the module are an ultrasound image with a point on the shaft and a point on the tip

Outputs to the module are a mask that can be used for supervised machine learning: Outputs to the module are a mask that can be used for supervised machine learning

TODO: Occasionally the VolumeClip does not fill the entire model: TODO: Occasionally the VolumeClip does not fill the entire model

Background and References