A Novel Tool for Supervised Segmentation Using 3D Slicer

A Novel Tool for Supervised Segmentation Using 3D Slicer

S S symmetry Article A Novel Tool for Supervised Segmentation Using 3D Slicer Daniel Chalupa * and Jan Mikulka Department of Theoretical and Experimental Electrical Engineering, Brno University of Technology, Technická 3082/12, 616 00 Brno, Czech Republic; [email protected] * Correspondence: [email protected] Received: 15 October 2018; Accepted: 7 November 2018; Published: 12 November 2018 Abstract: The rather impressive extension library of medical image-processing platform 3D Slicer lacks a wide range of machine-learning toolboxes. The authors have developed such a toolbox that incorporates commonly used machine-learning libraries. The extension uses a simple graphical user interface that allows the user to preprocess data, train a classifier, and use that classifier in common medical image-classification tasks, such as tumor staging or various anatomical segmentations without a deeper knowledge of the inner workings of the classifiers. A series of experiments were carried out to showcase the capabilities of the extension and quantify the symmetry between the physical characteristics of pathological tissues and the parameters of a classifying model. These experiments also include an analysis of the impact of training vector size and feature selection on the sensitivity and specificity of all included classifiers. The results indicate that training vector size can be minimized for all classifiers. Using the data from the Brain Tumor Segmentation Challenge, Random Forest appears to have the widest range of parameters that produce sufficiently accurate segmentations, while optimal Support Vector Machines’ training parameters are concentrated in a narrow feature space. Keywords: 3D slicer; classification; extension; random forest; segmentation; sensitivity analysis; support vector machine; tumor 1. Introduction 3D Slicer [1] is a free open-source platform for medical image visualization and processing. Its main functionality comes from the Extension Library, which consists of various modules that allow specific analyses of the input data, such as filtering, artefact suppression, or surface reconstruction. There is a lack of machine-learning extensions except for the open-source DeepInfer [2]. This deep-learning deployment kit uses 3D convolutional neural networks to detect and localize the target tissue. The development team demonstrated the use of this kit on the prostate segmentation problem for image-guided therapy. Researchers and practitioners are able to select a publicly available task-oriented network through the module without the need to design or train it. To enable the use of other machine-learning techniques, we developed the Supervised Segmentation Toolbox as an extensible machine-learning platform for 3D Slicer. Currently, Support Vector Machine (SVM) and Random Forest (RF) classifiers are included. These classifiers are well-researched and often used in image-processing tasks, as demonstrated in References [3–8]. SVMs [9] train by maximizing the distance between marginal samples (also referred to as support vectors) and a discriminative hyperplane by maximizing f in the equation: 1 f (a1 an) = ∑ ai ∑ ∑ aiajyiyj!xi !xj, (1) ··· − 2 i j · Symmetry 2018, 10, 627; doi:10.3390/sym10110627 www.mdpi.com/journal/symmetry SymmetrySymmetry20162018,,xx10,, x 627 22 of of 9 9 wherewhere yy isis defined defined as as +1 +1 for for a a class class A sample and –1 for a class class B sample, αa isis a a Lagrangian Lagrangian multiplier, multiplier, andand ~x!x isis the the feature feature vector vector of of the the individual individual sample. sample. On On real real data, data, this this is is often often too too strict strict because because ofof noisy noisy samples samples that that might might cross cross their their class class boundary. boundary. This This is is solved solved by by using using a a combination combination of of techniquestechniques known known as as the the kernel kernel trick trick and and soft soft margining. margining. The The kernel kernel trick trick uses uses a a kernel kernel function function φf toto remap remap the the original original feature feature space space to to a a higher-dimensional higher-dimensional one one by by replacing replacing the the dot dot product product in in EquationEquation ( (1)1) with withφf(x(x)) φf(xx ).. This This allows allows linear linear separation separation of of the the data data as required as required by the by SVM. the SVM. An i i ·· jj exampleAn example of the of the kernel kernel function function is the is the radial radial basis basis function function used used in in this this study. study. Incorporating Incorporating a a soft soft marginmargin allows allows some some samples samples to to cross cross their their class class boundary. boundary. Examples Examples of of such such soft-margin soft-margin SVMs SVMs are are the the C-SVMC-SVM [ [1010]] and and N-SVM N-SVM [ [1111].]. The The CC parameter of the C-SVM modifies the influence influence of of each each support support vectorvector on on the the final final discriminatory discriminatory hyperplane. hyperplane. The The larger larger the the CC,, the the closer closer the the soft-margin soft-margin C-SVM C-SVM is is to to aa hard-margin hard-margin SVM. SVM. The The N parameter of the N-SVM defines a minimum number of support vectors vectors and,and, consequently, consequently, an an upper upper bound bound of of the the guaranteed guaranteed maximum maximum percentage percentage of of misclassifications. misclassifications. FurtherFurther modifications modifications to to the the SVM SVM can can also also be be done, done, such such as as using using fuzzy-data fuzzy-data points points for for the the training training dataset,dataset, as as demonstrated demonstrated in in References References [ [55,,66].]. RFRF uses uses a a voting voting system system on on the the results results of of the the individual individual decision decision trees. trees. Each Each decision decision tree tree is is createdcreated on on the the basis basis of of a a bootstrapped bootstrapped sample sample of of the the training training data data [ [1212].]. 2.2. Materials Materials and and Methods Methods TheThe data data used used in in this this study study describe describe Low Low Grade Grade Glioma Glioma (LGG) (LGG) obtained obtained by by MRI. MRI. The The dataset dataset consistsconsists of of four four images images of of one one patient patient (LG_0001, (LG_0001, Figure Figure11):): T1- T1- and and T2-weighted, T2-weighted, contrast-enhanced contrast-enhanced T1C,T1C, and and Fluid Fluid Attenuated Attenuated Inversion Inversion Recovery Recovery (FL). (FL). These These data data are are part part of of the the training training set set featured featured in in thethe Brain Brain Tumor Tumor Segmentation Segmentation (BraTS) (BraTS) 2012 2012 Challenge. Challenge. FigureFigure 1. 1. SliceSlice 88 88 of of the the LG_0001 LG_0001 data data from from the the 2012 2012 Brain Brain Tumor Tumor Segmentation Segmentation (BraTS) (BraTS) Challenge. Challenge. TheThe dataset dataset consists consists of of a a ( (aa)) T1-weighted T1-weighted image image with with labels labels as as overlay; overlay; ( (bb)) postcontrast postcontrast T1-weighted T1-weighted image;image; ( (cc)) T2-weighted T2-weighted image; image; and and ( (dd)) Fluid Fluid Attenuated Attenuated Inversion Inversion Recovery Recovery (FL) (FL) image. image. TheThe free free and and open-source open-source Supervised Supervised Segmentation Segmentation Toolbox Toolbox extension extension [ [2413]] of of the the 3D 3D Slicer Slicer was was usedused throughout throughout this this study. study. The The extension extension allows allows the the user user to to train train a a range range of of classifiers classifiers using using labeled labeled data.data. The The user user is is also also able able to to perform perform a a grid grid search search to to select select the the optimal optimal parameters parameters for for the the classifier. classifier. ToTo achieve achieve this, this, the the extension extension uses uses either either an an already already available available function function or or a a cross-validation cross-validation algorithm algorithm developeddeveloped by by the the author author of of the the extension, extension, depending depending on on the the classifier classifier library library used. used. Currently, Currently, N-SVM N-SVM andand C-SVM C-SVM from from the the dlib dlib library library [ [1314]] and and C-SVM C-SVM and and Random Random Forest Forest from from Shark-ml Shark-ml library library [ [1415]] are are Symmetry 2016, xx, x 3 of 9 Symmetry 2018, 10, 627 3 of 9 incorporated.Symmetry 2016, xx, The x extension takes care of the parallelizable parts of the training and classification3 of 9 subtasks, thus significantly reducing computation times. A preprocessing algorithm selection is also a partincorporated. of the extension. The extension This allows takes for care artefact of the correction parallelizable or feature parts of extraction. the training The and extension classification workflow issubtasks, depicted thus in Figure significantly2. reducing reducing computation computation times. times. A A preprocessing preprocessing algorithm algorithm selection selection is is also also a apart part of of the the extension. extension. This This allows allows for for artefact artefact correction correction or or feature feature extraction. extraction. The The extension workflow is depicted in Figure2. Training Samples Preprocessing Training Segmentation Training Samples Preprocessing Training Segmentation Figure 2. Supervised segmentation extension workflow. 3. Results and DiscussionFigure 2. Supervised segmentation extension workflow. 3. ResultsA series and of Discussion tests were performed in order to provide a sense of the speed and accuracy of the provided classifiers. Sensitivity and specificity metrics were used to evaluate the results. A classifier A series of tests were performed in order to provide a sense of the speed and accuracy of the that had a larger sum of specificity and sensitivity (or their respective means, when cross-validation provided classifiers.

View Full Text

Details

  • File Type
    pdf
  • Upload Time
    -
  • Content Languages
    English
  • Upload User
    Anonymous/Not logged-in
  • File Pages
    9 Page
  • File Size
    -

Download

Channel Download Status
Express Download Enable

Copyright

We respect the copyrights and intellectual property rights of all users. All uploaded documents are either original works of the uploader or authorized works of the rightful owners.

  • Not to be reproduced or distributed without explicit permission.
  • Not used for commercial purposes outside of approved use cases.
  • Not used to infringe on the rights of the original creators.
  • If you believe any content infringes your copyright, please contact us immediately.

Support

For help with questions, suggestions, or problems, please contact us