Difference between revisions of "2016 Summer Project Week/Needle Segmentation from MRI"

From NAMIC Wiki
Jump to: navigation, search
 
(One intermediate revision by one other user not shown)
Line 2: Line 2:
 
<gallery>
 
<gallery>
 
Image:PW-Summer2016.png|[[2016_Summer_Project_Week#Projects|Projects List]]
 
Image:PW-Summer2016.png|[[2016_Summer_Project_Week#Projects|Projects List]]
 +
Image:CNN_pixel_by_pixel_tip_classification.png|[[2016_Summer_Project_Week#Projects|CNN pixel by pixel tip classification]]
 
</gallery>
 
</gallery>
  
Line 30: Line 31:
 
<h3>Progress</h3>
 
<h3>Progress</h3>
 
* GMM strategy failed.
 
* GMM strategy failed.
* CNN strategies has some accuracy problem but this can be the right way to get needles tips. In order to increase the accuracy some aspects can be investigated (patch size, dataset size, network architecture, data augmentation)
+
* We designed a pipeline to prepare the data for supervised learning.
 +
* We explored two strategies implementing Convolutional Neural Network.
 +
* CNN strategies have some accuracy problem but look promising. In order to increase the accuracy some aspects can be investigated such as patch size, dataset size, network architecture, data augmentation, MR data synthesising from CT. 
 
</div>
 
</div>
 
</div>
 
</div>

Latest revision as of 09:08, 25 June 2016

Home < 2016 Summer Project Week < Needle Segmentation from MRI

Key Investigators

  • Guillaume Pernelle, Imperial College London
  • Tina Kapur, BWH/HMS
  • Paolo Zaffino, ImagEngLab
  • Salvatore Scaramuzzino, ImagEngLab
  • Maria Francesca Spadea, ImagEngLab

Project Description

NeedleFinder offers tools to segment needles from MRI/CT. It has mostly been tested on MRI from GYN brachytherapy cases. Currently the user must provide the needle tip for the segmentation to start. We aim to detect the tip automatically to make the needle segmentation fully automatic.

Objective

  • Automatic detection of needle tips in MRI from GYN brachytherapy cases.

Approach, Plan

  • First possible strategy: to develop a detection method based on Gaussian Mixture Models (GMM) supplied with different inputs (intensity, Frangi's filter, needle-tip cross-correlation and Hough transform).
  • Second possible strategy: Convolutional Neural Network (CNN) over MRI patches. Two implementation can be investigated (pixel by pixel classification or entire patch classification).
  • Integrate algorithm to Slicer.

Progress

  • GMM strategy failed.
  • We designed a pipeline to prepare the data for supervised learning.
  • We explored two strategies implementing Convolutional Neural Network.
  • CNN strategies have some accuracy problem but look promising. In order to increase the accuracy some aspects can be investigated such as patch size, dataset size, network architecture, data augmentation, MR data synthesising from CT.