Tree-based Ensemble Classifier Learning for Automatic Brain Glioma Segmentation

Samya Amiri, Mohamed Ali Mahjoub, Islem Rekik

Research output: Contribution to journalArticle

3 Citations (Scopus)

Abstract

We introduce a dynamic multiscale tree (DMT) architecture that learns how to leverage the strengths of different state-of-the-art classifiers for supervised multi-label image segmentation. Unlike previous works that simply aggregate or cascade classifiers for addressing image segmentation and labeling tasks, we propose to embed strong classifiers into a tree structure that allows bi-directional flow of information between its classifier nodes to gradually improve their performances. Our DMT is a generic classification model that inherently embeds different cascades of classifiers while enhancing learning transfer between them to boost up their classification accuracies. Specifically, each node in our DMT can nest a Structured Random Forest (SRF) classifier or a Bayesian Network (BN) classifier. The proposed SRF-BN DMT architecture has several appealing properties. First, while SRF operates at a patch-level (regular image region), BN operates at the super-pixel level (irregular image region), thereby enabling the DMT to integrate multi-level image knowledge in the learning process. Second, although BN is powerful in modeling dependencies between image elements (superpixels, edges) and their features, the learning of its structure and parameters is challenging. On the other hand, SRF may fail to accurately detect very irregular object boundaries. The proposed DMT robustly overcomes these limitations for both classifiers through the ascending and descending flow of contextual information between each parent node and its children nodes. Third, we train DMT using different scales for input patches and superpixels. Basically, as we go deeper along the tree edges nearing its leaf nodes, we progressively decrease the patch and superpixel sizes, producing segmentation maps that capture a coarse-to-fine image details. Last, DMT demonstrates its outperformance in comparison to several state-of-the-art segmentation methods for multi-labeling of brain images with gliomas.
Original languageEnglish
Pages (from-to)135-142
Number of pages8
JournalNeurocomputing
Volume313
Early online date15 Jun 2018
DOIs
Publication statusPublished - 3 Nov 2018

Fingerprint

Glioma
Brain
Classifiers
Learning
Bayesian networks
Image segmentation
Labeling
Labels
Pixels

Keywords

  • Boosting
  • Brain
  • Dynamic tree
  • Ensemble classifier learning
  • Segmentation
  • Tumor

Cite this

Amiri, Samya ; Ali Mahjoub, Mohamed ; Rekik, Islem. / Tree-based Ensemble Classifier Learning for Automatic Brain Glioma Segmentation. In: Neurocomputing. 2018 ; Vol. 313. pp. 135-142.
@article{382d2f131b8d442aaccee1242eda104d,
title = "Tree-based Ensemble Classifier Learning for Automatic Brain Glioma Segmentation",
abstract = "We introduce a dynamic multiscale tree (DMT) architecture that learns how to leverage the strengths of different state-of-the-art classifiers for supervised multi-label image segmentation. Unlike previous works that simply aggregate or cascade classifiers for addressing image segmentation and labeling tasks, we propose to embed strong classifiers into a tree structure that allows bi-directional flow of information between its classifier nodes to gradually improve their performances. Our DMT is a generic classification model that inherently embeds different cascades of classifiers while enhancing learning transfer between them to boost up their classification accuracies. Specifically, each node in our DMT can nest a Structured Random Forest (SRF) classifier or a Bayesian Network (BN) classifier. The proposed SRF-BN DMT architecture has several appealing properties. First, while SRF operates at a patch-level (regular image region), BN operates at the super-pixel level (irregular image region), thereby enabling the DMT to integrate multi-level image knowledge in the learning process. Second, although BN is powerful in modeling dependencies between image elements (superpixels, edges) and their features, the learning of its structure and parameters is challenging. On the other hand, SRF may fail to accurately detect very irregular object boundaries. The proposed DMT robustly overcomes these limitations for both classifiers through the ascending and descending flow of contextual information between each parent node and its children nodes. Third, we train DMT using different scales for input patches and superpixels. Basically, as we go deeper along the tree edges nearing its leaf nodes, we progressively decrease the patch and superpixel sizes, producing segmentation maps that capture a coarse-to-fine image details. Last, DMT demonstrates its outperformance in comparison to several state-of-the-art segmentation methods for multi-labeling of brain images with gliomas.",
keywords = "Boosting, Brain, Dynamic tree, Ensemble classifier learning, Segmentation, Tumor",
author = "Samya Amiri and {Ali Mahjoub}, Mohamed and Islem Rekik",
year = "2018",
month = "11",
day = "3",
doi = "10.1016/j.neucom.2018.05.112",
language = "English",
volume = "313",
pages = "135--142",
journal = "Neurocomputing",
issn = "0925-2312",
publisher = "Elsevier",

}

Tree-based Ensemble Classifier Learning for Automatic Brain Glioma Segmentation. / Amiri, Samya; Ali Mahjoub, Mohamed; Rekik, Islem.

In: Neurocomputing, Vol. 313, 03.11.2018, p. 135-142.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Tree-based Ensemble Classifier Learning for Automatic Brain Glioma Segmentation

AU - Amiri, Samya

AU - Ali Mahjoub, Mohamed

AU - Rekik, Islem

PY - 2018/11/3

Y1 - 2018/11/3

N2 - We introduce a dynamic multiscale tree (DMT) architecture that learns how to leverage the strengths of different state-of-the-art classifiers for supervised multi-label image segmentation. Unlike previous works that simply aggregate or cascade classifiers for addressing image segmentation and labeling tasks, we propose to embed strong classifiers into a tree structure that allows bi-directional flow of information between its classifier nodes to gradually improve their performances. Our DMT is a generic classification model that inherently embeds different cascades of classifiers while enhancing learning transfer between them to boost up their classification accuracies. Specifically, each node in our DMT can nest a Structured Random Forest (SRF) classifier or a Bayesian Network (BN) classifier. The proposed SRF-BN DMT architecture has several appealing properties. First, while SRF operates at a patch-level (regular image region), BN operates at the super-pixel level (irregular image region), thereby enabling the DMT to integrate multi-level image knowledge in the learning process. Second, although BN is powerful in modeling dependencies between image elements (superpixels, edges) and their features, the learning of its structure and parameters is challenging. On the other hand, SRF may fail to accurately detect very irregular object boundaries. The proposed DMT robustly overcomes these limitations for both classifiers through the ascending and descending flow of contextual information between each parent node and its children nodes. Third, we train DMT using different scales for input patches and superpixels. Basically, as we go deeper along the tree edges nearing its leaf nodes, we progressively decrease the patch and superpixel sizes, producing segmentation maps that capture a coarse-to-fine image details. Last, DMT demonstrates its outperformance in comparison to several state-of-the-art segmentation methods for multi-labeling of brain images with gliomas.

AB - We introduce a dynamic multiscale tree (DMT) architecture that learns how to leverage the strengths of different state-of-the-art classifiers for supervised multi-label image segmentation. Unlike previous works that simply aggregate or cascade classifiers for addressing image segmentation and labeling tasks, we propose to embed strong classifiers into a tree structure that allows bi-directional flow of information between its classifier nodes to gradually improve their performances. Our DMT is a generic classification model that inherently embeds different cascades of classifiers while enhancing learning transfer between them to boost up their classification accuracies. Specifically, each node in our DMT can nest a Structured Random Forest (SRF) classifier or a Bayesian Network (BN) classifier. The proposed SRF-BN DMT architecture has several appealing properties. First, while SRF operates at a patch-level (regular image region), BN operates at the super-pixel level (irregular image region), thereby enabling the DMT to integrate multi-level image knowledge in the learning process. Second, although BN is powerful in modeling dependencies between image elements (superpixels, edges) and their features, the learning of its structure and parameters is challenging. On the other hand, SRF may fail to accurately detect very irregular object boundaries. The proposed DMT robustly overcomes these limitations for both classifiers through the ascending and descending flow of contextual information between each parent node and its children nodes. Third, we train DMT using different scales for input patches and superpixels. Basically, as we go deeper along the tree edges nearing its leaf nodes, we progressively decrease the patch and superpixel sizes, producing segmentation maps that capture a coarse-to-fine image details. Last, DMT demonstrates its outperformance in comparison to several state-of-the-art segmentation methods for multi-labeling of brain images with gliomas.

KW - Boosting

KW - Brain

KW - Dynamic tree

KW - Ensemble classifier learning

KW - Segmentation

KW - Tumor

UR - http://www.scopus.com/inward/record.url?scp=85049439057&partnerID=8YFLogxK

U2 - 10.1016/j.neucom.2018.05.112

DO - 10.1016/j.neucom.2018.05.112

M3 - Article

VL - 313

SP - 135

EP - 142

JO - Neurocomputing

JF - Neurocomputing

SN - 0925-2312

ER -