Semi-automated brain tumor and edema segmentation using MRI

doi:10.1016/j.ejrad.2005.03.028

European Journal of Radiology

Volume 56, Issue 1, October 2005, Pages 12-19

https://doi.org/10.1016/j.ejrad.2005.03.028 Get rights and content

Abstract

Purpose:

Manual segmentation of brain tumors from magnetic resonance images is a challenging and time-consuming task. A semi-automated method has been developed for brain tumor and edema segmentation that will provide objective, reproducible segmentations that are close to the manual results. Additionally, the method segments non-enhancing brain tumor and edema from healthy tissues in magnetic resonance images.

Methods and materials:

In this study, a semi-automated method was developed for brain tumor and edema segmentation and volume measurement using magnetic resonance imaging (MRI). Some novel algorithms for tumor segmentation from MRI were integrated in this medical diagnosis system. We exploit a hybrid level set (HLS) segmentation method driven by region and boundary information simultaneously, region information serves as a propagation force which is robust and boundary information serves as a stopping functional which is accurate. Ten different patients with brain tumors of different size, shape and location were selected, a total of 246 axial tumor-containing slices obtained from 10 patients were used to evaluate the effectiveness of segmentation methods.

Results:

This method was applied to 10 non-enhancing brain tumors and satisfactory results were achieved. Two quantitative measures for tumor segmentation quality estimation, namely, correspondence ratio (CR) and percent matching (PM), were performed. For the segmentation of brain tumor, the volume total PM varies from 79.12 to 93.25% with the mean of 85.67 ± 4.38% while the volume total CR varies from 0.74 to 0.91 with the mean of 0.84 ± 0.07. For the segmentation of edema, the volume total PM varies from 72.86 to 87.29% with the mean of 79.54 ± 4.18% while the volume total CR varies from 0.69 to 0.85 with the mean of 0.79 ± 0.08. The HLS segmentation method perform better than the classical level sets (LS) segmentation method in PM and CR.

Conclusions:

The results of this research may have potential applications, both as a staging procedure and a method of evaluating tumor response during treatment, this method can be used as a clinical image analysis tool for doctors or radiologists.

Introduction

The task of manually segmenting brain tumors from magnetic resonance imaging (MRI) is generally time-consuming and difficult. In most settings, the task is performed by marking the tumor regions slice-by-slice, which limits the human rater's view and generates jaggy images as in Fig. 1. As a result, the segmented images are less than optimal. A semi-automated segmentation method is desirable because it reduces the load on the human raters and generates satisfactory segmentation results.

Tumor volume is a significant prognostic factor in the treatment of malignant tumors. Treatment protocols for malignant brain tumors generally call for removal through surgical procedures followed by irradiation of the tumor bed. The goal of three-dimensional (3-D) conformal radiation therapy is to irradiate the tumor volume while limiting damage to the surrounding normal tissues. Achieving this goal requires accurate determination of 3-D treatment volumes.

Although the technology for conformal radiation treatment planning has developed to a high level of accuracy, the definition of the gross tumor volume (GTV) in MRI is still based on time-intensive, highly subjective manual outlining [1], [2], [3]. Manual outlining is the type of process that should be an excellent candidate for automation through the development of a computerized segmentation system. At our institution, several techniques of MRI segmentation have been developed and evaluated specifically for brain tumors.

Tumor volume is a significant prognostic factor in the treatment of malignant tumors [4], [5], [6], [7]. Multivariate modeling has shown that tumor volume is a dominant covariate that overwhelms T stage, N stage and stage group [1], [8], [9]. The TNM classification system recognizes the importance of tumor volume. This article describes the classical level sets (LS) and hybrid level set (HLS) semi-automated methods for brain tumor segmentation and volume measurement. The results of this research may have potential applications, both as a staging procedure and a method of evaluating tumor response during treatment.

Section snippets

Tumor characteristics and assumptions

Brain tumors are difficult to segment because they have a wide range of appearance and effect on surrounding structures. Following are some of the general characteristics of brain tumors: (A) vary greatly in size and position, (B) vary greatly in the way they show up in MRI, (C) may have overlapping intensities with normal tissue, (D) may be space occupying (new tissue that moves normal structure) or infiltrating (changing properties of existing tissue), (E) may enhance fully, partially, or not

Phantom validation

Two hundred and forty six slices of the phantom were processed using those three methods mentioned above. The results are shown in Table 1. The errors of LS and HLS were 3.8–7.4 and 3.6–6.5%, respectively, whereas the error of manual tracing and labeling was 3.5–7.8%.

Brain tumor volume

A total of 246 axial tumor-containing slices obtained from 10 patients were evaluated using manual tracing and semi-automated segmentation methods. The results of tumor volume measurement are presented in Table 2, Table 3. There

Discussion

Segmentation is very important in this system. Some clinically or research available visualization systems had been developed by other researchers and among them, one for head and neck using CT [10], one for liver using CT [11] and two for brain using MRI [12], [13]. For these systems, the pathology segmentation or feature extraction might not be very complex because: (1) for CT, the image intensity distribution feature is simple and (2) for brain, the image content feature is simple. However,

Conclusions

Tumor volume is an important diagnostic indicator in treatment planning and results assessment for brain tumor, which has a high frequency in South China. The measurement of brain tumor volume could assist tumor staging and effective treatment planning.

Imaging plays a central role in the diagnosis and treatment planning of brain tumor. In this study, a semi-automated system for brain tumor volume measurements was developed based on MRI. Due to the image and anatomic features, a HLS method for

Acknowledgements

This work was partly supported by the Programme de Recherches Avancées de Coopérations Franco-Chinoises (PRA SI 03-03) and the Region Rhône-Alpes of France within the project “MIRA Recherche 2003”.

References (13)

D.J. Brenner
Dose, volume and tumor control predictions in radiotherapy
Int J Radiat Oncol Biol Phys
(1993)
D.J. Brenner
Dose, volume and tumor control predictions in radiotherapy
Int J Radiat Oncol Biol Phys
(1993)
C.R. Johnson et al.
The tumor volume and clonogen number relationship: Tumor control predictions based upon tumor volume estimates derived from computed tomography
Int J Radiat Oncol Biol Phys
(1995)
D.T. Chua et al.
Volumetric analysis of tumor extent in nasopharyngeal carcinoma and correlation with treatment outcome
Int J Radiat Oncol Biol Phys
(1997)
C.R. Johnson et al.
The influence of quantitative tumor volume measurements on local control in advanced head and neck cancer using concomitant boost accelerated superfractionated irradiation
Int J Radiat Oncol Biol Phys
(1995)
J.S. Lee et al.
Volumetric visualization of head and neck CT data for treatment planning
Int J Radiat Oncol Biol Phys
(1999)

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

Cited by (106)

Brain MRI segmentation using deep learning: Background study and challenges
2021, Brain Tumor MRI Image Segmentation Using Deep Learning Techniques
The segmentation of brain tumors is an important component in medical image analysis. Early detection of brain tumors improves diagnostic methods and increases patients' chances of survival. Manual segmentation of brain tumors from large amounts of magnetic resonance imaging (MRI) images produced in clinical routine is a complicated and time-consuming process. Thus, an automatic way to brain tumor image segmentation is needed. This chapter aims to deliver an overview of deep learning (DL) based brain MRI segmentation. Automatic segmentation utilizing DL methods has recently gained popularity as these methods produce the state-of-the-art performance and can solve this issue better than the traditional. There are a variety of review papers available that concentrate on conventional approaches for MRI-based brain tumor image segmentation. Unlike others, the emphasis in this chapter is on the current developments of DL methods in this area, as well as the challenges and future aspects. First, an overview of brain tumors and segmentation methods is given. The state-of-the-art architecture is then addressed, with an emphasis on the current developments of DL methods. Finally, an overview of the challenges is discussed in the field of MRI-based brain tumor segmentation approaches into everyday clinical practice.
Brain tumor classification based on DWT fusion of MRI sequences using convolutional neural network
2020, Pattern Recognition Letters
Tumor in brain is an anthology of anomalous cells. It leads to increase in death rate among humans. Therefore, in this manuscript, a fusion process is proposed to combine structural and texture information of four MRI sequences (T1C, T1, Flair and T2) for the detection of brain tumor. A discrete wavelet transform (DWT) along with Daubechies wavelet kernel is utilized for fusion process which provides a more informative tumor region as compared to an individual single sequence of MRI. After the fusion process, a partial differential diffusion filter (PDDF) is applied to remove noise. A global thresholding method is used for segmenting tumor region which is then fed to proposed convolutional neural network (CNN) model for finally differentiating tumor and non-tumor regions. Five publicly available datasets i.e., BRATS 2012, BRATS 2013, BRATS 2015, BRATS 2013 Leader board and BRATS 2018 are used for proposed method evaluation. The results show that fused images provide better results as compared to individual sequences on benchmark datasets.
Improved model prediction of glioma growth utilizing tissue-specific boundary effects
2019, Mathematical Biosciences
Kinetic parameter estimates for mathematical models of glioblastoma multiforme (GBM), derived from clinical scans, have been used to predict the occurrence of hypoxia, necrosis, response to radiation therapy, and overall survival. Modeling GBM growth in a cerebral model encounters anatomical boundaries that interfere with model calibration from clinical measurements.
Methods: The effect of boundaries is examined on both spherically symmetric and anatomical models of tumor growth. This effect is incorporated into a method that updates kinetic parameters. The efficacy of this method in reproducing clinical image-derived subject data is evaluated.
Results: Spherically symmetric simulations of tumor growth with simple boundaries behave predictably when in a linear phase of growth. Anatomic simulations of eleven out of twenty subjects demonstrated improved fit to subject data with the new method. When only subjects exhibiting linear growth are considered, eight out of nine subject demonstrate improved fit to the data.
Conclusion: Anatomical boundaries to tumor growth measurably deflect progression and affect estimates of kinetic parameters. The presented method reliably updates kinetic parameters to fit anatomic computational models to clinically derived subject data when those data are in a linear regime.
Self-adaptive dragonfly based optimal thresholding for multilevel segmentation of digital images
2018, Journal of King Saud University - Computer and Information Sciences
Dragonfly optimization (DFO) is a population based meta-heuristic optimization algorithm that simulates the static and dynamic swarming behaviors of dragonflies. The static swarm comprising less number of dragonflies in a small area for hunting preys, while the dynamic swarm with a large number of dragonflies migrates over long distances; and they represent the exploration and exploitation phases of the DFO. This paper introduces a self adaptive scheme for tuning the DFO parameters and suggests a methodology involving self-adaptive DFO (SADFO) for performing multilevel segmentation of digital images. The multilevel segmentation problem is formulated as an optimization problem and solved using the SADFO. The method optimizes the threshold values through effectively exploring the solution space in obtaining the global best solution. The results of real life and medical images illustrate the performance of the suggested method.
On differentiation between vasogenic edema and non-enhancing tumor in high-grade glioma patients using a support vector machine classifier based upon pre and post-surgery MRI images
2018, European Journal of Radiology
Citation Excerpt :
Deep learning is another widely used supervised machine learning approach [17], but it typically requires a large amount of data for training purposes. There has been considerable research in brain tumor segmentation using conventional images such as T1-W, T2-W, and FLAIR [18,19], as well as combining them with other MRI modalities such as T1-W contrast enhanced images, diffusion-weighted images, and spectroscopy-based techniques [20–22]. Most of these studies have relied on supervised tumor segmentation methods where the ground truth for tumor subparts was delineated by neuroradiologists [3,8].
High grade gliomas (HGGs) are infiltrative in nature. Differentiation between vasogenic edema and non-contrast enhancing tumor is difficult as both appear hyperintense in T₂-W/FLAIR images. Most studies involving differentiation between vasogenic edema and non-enhancing tumor consider radiologist-based tumor delineation as the ground truth. However, analysis by a radiologist can be subjective and there remain both inter- and intra-rater differences. The objective of the current study is to develop a methodology for differentiation between non-enhancing tumor and vasogenic edema in HGG patients based on T₁ perfusion MRI parameters, using a ground truth which is independent of a radiologist’s manual delineation of the tumor.
This study included 9 HGG patients with pre- and post-surgery MRI data and 9 metastasis patients with pre-surgery MRI data. MRI data included conventional T₁-W, T₂-W, and FLAIR images and DCE-MRI dynamic images. In this study, the authors hypothesize that surgeried non-enhancing FLAIR hyperintense tissue, which was obtained using pre- and post-surgery MRI images of glioma patients, should be largely comprised of non-enhancing tumor. Hence this could be used as an alternative ground truth for the non-enhancing tumor region. Histological examination of the resected tissue was done for validation. Vasogenic edema was obtained from the non-enhancing FLAIR hyperintense region of metastasis patients, as they have a clear boundary between enhancing tumor and edema. DCE-MRI data analysis was performed to obtain T₁ perfusion MRI parameters. Support Vector Machine (SVM) classification was performed using T₁ perfusion MRI parameters to differentiate between non-enhancing tumor and vasogenic edema. Receiver-operating-characteristic (ROC) analysis was done on the results of the SVM classifier. For improved classification accuracy, the SVM output was post-processed via neighborhood smoothing.
Histology results showed that resected tissue consists largely of tumorous tissue with 7.21 ± 4.05% edema and a small amount of healthy tissue. SVM-based classification provided a misclassification error of 8.4% in differentiation between non-enhancing tumor and vasogenic edema, which was further reduced to 2.4% using neighborhood smoothing.
The current study proposes a semiautomatic method for segmentation between non-enhancing tumor and vasogenic edema in HGG patients, based on an SVM classifier trained on an alternative ground truth to a radiologist’s manual delineation of a tumor. The proposed methodology may prove to be a useful tool for pre- and post-operative evaluation of glioma patients.
Analyzing the Performance of SVM-ACO Classifier and Hybrid Optimization Techniques in MRI Brain Tumor Segmentation for Early Prognosis
2024, International Journal of Intelligent Systems and Applications in Engineering

View all citing articles on Scopus

View full text

Semi-automated brain tumor and edema segmentation using MRI

Abstract

Purpose:

Methods and materials:

Results:

Conclusions:

Introduction

Section snippets

Tumor characteristics and assumptions

Phantom validation

Brain tumor volume

Discussion

Conclusions

Acknowledgements

Int J Radiat Oncol Biol Phys

Int J Radiat Oncol Biol Phys

Int J Radiat Oncol Biol Phys

Int J Radiat Oncol Biol Phys

Int J Radiat Oncol Biol Phys

Int J Radiat Oncol Biol Phys