Elsevier

Pattern Recognition Letters

Volume 46, 1 September 2014, Pages 1-10
Pattern Recognition Letters

Iterative multi-class multi-scale stacked sequential learning: Definition and application to medical volume segmentation

https://doi.org/10.1016/j.patrec.2014.05.002Get rights and content

Highlights

  • We present the iterative multiclass multiscale stacked sequential learning framework.

  • This method learns a set of contextual properties from a stack of learners.

  • We test the method on three medical volume multi-class segmentation problems.

  • The method is easy to implement and independent to the feature space and classifier.

Abstract

In this work we present the iterative multi-class multi-scale stacked sequential learning framework (IMMSSL), a novel learning scheme that is particularly suited for medical volume segmentation applications. This model exploits the inherent voxel contextual information of the structures of interest in order to improve its segmentation performance results. Without any feature set or learning algorithm prior assumption, the proposed scheme directly seeks to learn the contextual properties of a region from the predicted classifications of previous classifiers within an iterative scheme. Performance results regarding segmentation accuracy in three two-class and multi-class medical volume datasets show a significant improvement with respect to state of the art alternatives. Due to its easiness of implementation and its independence of feature space and learning algorithm, the presented machine learning framework could be taken into consideration as a first choice in complex volume segmentation scenarios.

Section snippets

Introduction and related work

Machine learning schemes have been widely applied in many image and volume segmentation scenarios [1], [2]. A common property of image and volume data is the high context dependence condition of any pixel or voxel value, which breaks the common assumptions of most machine learning strategies regarding statistical independence of the entries. Although this fact could be ignored during the machine learning framework design, a number of works have tried to include it, showing a significant

The IMMSSL framework

In this section, we present the iterative multiclass multiscale stacked sequential learning scheme for volume segmentation. First, we review the stacked sequential learning (SSL) [16] and the multiclass multiscale stacked sequential learning (MMSSL) [17], [18] schemes. Then, an improved iterative framework built on top of them in the context of 3D segmentation applications is presented (IMMSL).

IMMSSL application to medical volume segmentation problems

In this section, we evaluate the performance of the proposed IMMSSL framework in three medical volume segmentation problems. The first one is a two-class problem where we are interested to detect and segment any tumor volume present within a whole body Positron Emission Tomography (PET) scan. The second one aims to segment a brain volume model (www.slicer.org/archives) in three compartments (right hemisphere, cerebellum and left hemisphere) providing as input only a stratified small percentage

Conclusions

In this work an Iterative 3D-scale generalization of the multiclass multiscale stacked sequential learning framework has been presented. The method accurately models the high context-dependence of the voxel properties in volume segmentation problems by introducing an iterative learning scheme of stacked contextual classifiers. The performance of the method has been tested on different medical volume problems containing a high variety of structures to segment. Its overall performance results and

Acknowledgments

The work of Frederic Sampedro is supported by the Spanish government FPU (Formación del Profesorado Universitario) doctoral grant. The authors would like to express their great appreciation to Dr Ignasi Carrio, director of the nuclear medicine department of the Hospital de Sant Pau. Work partially funded by TIN2011-24220 Spanish projects.

References (21)

There are more references available in the full text version of this article.

Cited by (5)

  • A new approach to contextual learning using interval arithmetic and its applications for land-use classification

    2016, Pattern Recognition Letters
    Citation Excerpt :

    Afterwards, Puertas et al. [19] addressed the aforementioned work in the context of multi-class-based classification problems. Sampedro et al. [21] proposed a similar approach to that one introduced by Puertas et al. [19], but now in a three-dimensional space, which has been used together with error-correcting output codes in the context of medical image classification. An interesting approach is to design hybrid versions of well-known classifiers in order to consider contextual information by means of Markov Random Fields (MRFs).

  • Segmentation of RGB-D indoor scenes by stacking random forests and conditional random fields

    2016, Pattern Recognition Letters
    Citation Excerpt :

    Instead of a single distance of contextual information, as used in [3], multiple distances are used which outperforms previous methods. Also in [16], classifiers are stacked iteratively until some stopping criteria. They call it the iterative multi-class multi-scale stacked sequential learning or IMMSSL.

  • An Ensemble-Based Stacked Sequential Learning Algorithm for Remote Sensing Imagery Classification

    2017, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing

This paper has been recommended for acceptance by C. Luengo.

View full text