Minimally parametrized segmentation framework with dual metaheuristic optimisation algorithms and FCM for detection of anomalies in MR brain images
Graphical abstract
Introduction
Many radiologists say it's very challenging to diagnose brain-related problems; particularly tumour infiltration. Owing to the complex structure of the brain, it is also hard to distinguish the tumour-affected region and the edema region that surrounds it. Hence, oncologists and radiologists need an advanced automated diagnosing system to improve image visualisation for precise anomaly detection and infectious tissue fragmentation. Clinical experts prefer Magnetic Resonance Imaging (MRI) for diagnosis because it provides more information than other imaging modalities such as Computed Tomography (CT), X-Ray Imaging, Positron Emission Tomography (PET), and Ultrasound. However, MR Images may get distorted due to the movement of the patient during the scanning process and this challenge can be mitigated with the usage of a special sensing device mentioned in [49].
Manual segmentation takes a long processing time and can produce inconsistent findings due to intra-observer and inter-observer variability. Using a fully automated segmentation methodology with appropriate graphical tools, it is possible to segment brain lesions and other tissue regions into numerous separate parts without the need for human interaction. Many automatic segmentation [1], [37] algorithms have been developed in previous studies to comprehend tumour characteristics, like volume, size, shape, type and location. However, selecting an appropriate segmentation algorithm is very important to obtain the desired results. Although many optimisation algorithms have been developed in the literature, most of them are biased by the algorithm-specific parameters and may lead to inferior segmentation. To overcome the abovementioned problems, two minimally parameterized optimisation algorithms, namely, the Artificial Bee Colony and the novel JAYA algorithm, have been recommended in the study. The first stage optimisation enhances the segmented region based on the texture feature from GLCM, and the second stage optimises the output with a suitable modification in population size.
In general, five basic techniques are developed based on predefined attributes such as Edge, Region Growing, Threshold, Neural Network, and Fuzzy Logic for the fully automated image segmentation process. In the edge-based methods [2], Gradient (Sobel, Canny), Gaussian (Laplacian) based operators are used as edge-filter for the detection of dissimilarity in an image. In this method, pixels are categorized into the edge (converging to one) or non-edge (converging to zero), and thereby a target boundary is constructed. The accuracy level of this method depletes with the increase in homogeneousness (similarity in grayscale level distribution) between CSF, WM, and GM. Consequently, region-growing methods [3] have also been preferred for the tumour segmentation process in which a feasible solution is obtained for inhomogeneous brain tissues. In this method, if the nearby pixels obey the predefined intensity level guidelines, then the pixel is included in the region, and the procedure continues till all the pixels in the image are checked for similarity. The key limitation of the region-growing method is its high dependency on the order of scanning pixels. Different scanning positions may offer different segmentation outputs. In the threshold-based methods [4], the tumour region is isolated from the background by setting an appropriate threshold limit, which directly influences the sensitivity and selectivity property of the segmentation. In literature, various thresholding techniques were employed for image segmentation such as Otsu [5], maximum correlation, and optimal thresholding. Even a small variation in the threshold limit may impact the segmentation output adversely. Active Contour [5] model is considered widely for image segmentation due to its ability to delineate the region of interest from the background. However, this method needs a long duration to track the entire path from initial contour points.
In recent times, most researchers have preferred Artificial Neural Network (ANN) [6] based segmentation for producing robust results with less computational time at a moderate cost, backed by its different set of training parameters. ANN-based methods ensure higher efficacy when many slices are employed for the segmentation process. Though ANN has many merits in terms of multitasking capability, operational speed, and accuracy, it also has its demerits like unknown operational behaviour and selection of appropriate network structure. Many toolbox-based courseware were developed in the literature to ease the segregation of data. In [51], the authors developed the ADDIE methodology using a toolbox for the right identification and isolation of COVID-19 patients among the hajj pilgrims. Minimization of the number of instructions fed to the toolbox instructor can be achieved to abstain from human-driven error inputs. The fuzzy logic [7] is frequently used in image segmentation process due to its easy accessibility to large datasets and its ability to segment complex tissue structures. FCM clustering is utilized in this proposed methodology due to its salient features: (i) Membership value is assigned to data elements based on the fuzzy logic set, (ii) Allows data element to belong to multiple clusters if it satisfies the criterion of more than one clusters, (iii) Minimize the objective function in fewer iteration. Fuzzy-based clustering algorithms [8] are the best segmentation tool when there is a lot of vagueness and complex structures in the MR brain image. In literature, many FCM based (derived/modified) methodologies were developed such as Modified Kernel FCM [9], Fast Spatial FCM (FSFCM) [10], Type-2 adaptive weighted spatial FCM (AWSFCM) [11], Bat Algorithm with Fuzzy C-Ordered Means (BAFCOM) [12], K-means and modified fuzzy C-means [13] etc.
Feature extraction [50] is the method of grouping data points with the same attributes as frequency, pixel intensity, texture, and colour to reduce the dimensionality of the input test image for further processing. Choosing the appropriate traits from the test image is significant because the incorrect selection of attributes may degrade the overall segmentation results. In the proposed algorithm, the GLCM was utilized [14], [15] due to its effectiveness in redefining the image with reduced dimension (set of texture features) using its gray-level intensity.
Numerous optimisation techniques were deployed in the literature to enhance the segmentation accuracy, such as Evolutionary Algorithms [16], [17] and Swarm Intelligence Algorithms [18], [19], [20], [21], [22].
Unlike other optimisation algorithms, ABC requires only a minimal number of algorithm-specific parameters. For instance, a standard Genetic Algorithm needs three major control parameters, such as crossover mutation, mutation rate, and generation gap, apart from the population size and maximum evaluation number. Particle Swarm Optimisation (PSO) algorithm also needs control parameters like cognitive, acceleration coefficient, inertia weight, and particle count apart from the maximum velocity. On the other hand, the ABC algorithm needs only one control parameter in addition to the two algorithm-specific parameters like the number of food sources and maximum search count. Secondly, the ABC algorithm is easy and simple to hybridize with other algorithms and the third advantage is its fast convergence offered. Owing to the above-mentioned salient features, the authors have chosen the Artificial Bee Colony [23] algorithm for the initial level of optimisation.
Though the ABC algorithm has merits like the ease of implementation and the ability to explore local solutions with a fast convergence rate, it also has its demerits in premature convergence during the search period. In many cases, single-stage optimisation could not complete the search target owing to algorithm-specific limitations, which can be resolved by developing a two-stage optimisation [34] to examine the brain abnormality in MR brain images. To overcome the above-said dissensions, the JAYA algorithm [24] was used as the second optimisation technique to enhance segmented output further. It has the following advantages, such as (i) parameter less tuning, (ii) taking very little iteration to minimize the objective function, and (iii) requires very less memory to process voluminous slices. In the study, a hybrid segmentation framework was developed by improvising the métiers and reducing the weaknesses influenced by each algorithm to detect tumour regions effectually.
It is well known that no optimisation technique is superior to the other(s) in solving all optimisation related problems. However, literature shows that an optimisation technique may address a specific problem effectively. On the back of this, an integrated metaheuristic algorithm, a combination of the improved ABC algorithm and the novel JAYA algorithm, was developed in the research. The motivations for this study are.
- •
To develop a fully automatic system with dual metaheuristic optimisation techniques to segment brain MRI with a higher success rate than the single optimisation algorithm,
- •
To address the exploitation and exploration limitations confronted by the optimisation algorithms during search operation with suitable modification to its standard operations, and
- •
To reduce the overall convergence computation cost by using the fewest possible control parameters.
To achieve the above-mentioned motivations, the contributions made to this research are as follows:
- •
To the best of knowledge, this is the first work which integrates two metaheuristic optimisation algorithms at different stages to enhance the segmentation output with minimal algorithm specific control parameters, as shown in Fig. 1.
- •
For the first time in literature, GLCM features were used to enhance the ABC algorithm’s exploitation capability during its search operation.
- •
First and foremost, in literature survey made, the novel JAYA algorithm with two improvements to its standard operations is introduced in the spatial domain for re-evaluation of the segmented image.
- •
Assessing the outcomes in both the normal and noise-embedded images to show the ability of the proposed methodology.
The paper's organization as follows, section 2 elaborates the material used and methodology of the recommended algorithm. In section 3, a detailed explanation of the proposed novel Dual Optimisation Algorithm (DOA) and their respective pseudocodes is discussed. Section 4 briefs results and compares the proposed algorithm's performance to other competing methods available in the literature. Lastly, the conclusion and future work planned are presented in section 5.
Section snippets
Proposed methodology
Fig. 1 explains the functioning of the novel Dual Optimisation algorithm. Firstly, the MR brain image is filtered using a Gaussian filter to remove noise. Then, the filtered image is segmented using the FCM clustering technique with 4 cluster heads. The Gray Level Co-occurrence Matrix is utilized to extract texture features from the segmented image to ascertain the optimal (minimum and maximum) threshold values during initial optimisation.
Then, the segmented image further undergoes the first
Preprocessing using Gaussian filter
Generally, test images need to be pre-processed before segmentation due to their high vulnerability to noise. In this work, a Gaussian filter was utilized to reduce the noise signal that appears in the MR test image. A combination of T1 weighted, T2 weighted, and FLAIR (Fluid-Attenuated Inversion Recovery) images were used as input for the proposed algorithm [36]. It is known that the Gray tone level of tumour and edema portion is low for T1 and FLAIR types of images, whereas the same is high
Results and Discussion
This section explores the MR brain images carrying different tumour complaints qualitatively and quantitatively. The proposed novel DOA framework effectively identifies the tumour region by differentiating the various tissue parts. The performance of FCM based JA-ABC algorithm is comprehensively compared with standalone FCM and JA-FCM algorithms to understand the progressive improvement achieved by the recommended novel DOA toward tumour region identifiaction.
About 25 clinical brain images with
Conclusion
The proposed framework based on JA-ABC FCM has been tested on a set of MR brain images and it is much better on comparison with the conventional metaheuristic approaches. The obtained results have been compared to the various standalone, consolidated algorithms and recently developed metaheuristic algorithms and machine learning methods. By virtue of superior visual comparison along with key performance assessment metrics, it is proved that the suggested algorithm produces promising results
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
References (52)
- et al.
Brain tumour segmentation based on a new threshold approach
Procedia Comput. Sci.
(2017) - et al.
An adaptive differential evolution algorithm to optimal multi-level thresholding for MRI brain image segmentation
Expert Syst. Applications
(2019) Optimization of methods for image-texture segmentation using ant colony optimization
- et al.
A medical analytical system using intelligent fuzzy level set brain image segmentation based on improved quantum particle swarm optimization
Microprocess. Microsyst.
(2020) - et al.
An integrated design of particle swarm optimization (PSO) with fusion of features for detection of brain tumor
Pattern Recogn. Lett.
(2020) - et al.
Teaching–learning-based optimization: a novel method for constrained mechanical design optimization problems
Comput.-Aided Design
(2011) - et al.
Development of a combinational framework to concurrently perform tissue segmentation and tumor identification in T1-W, T2-W, FLAIR and MPR type magnetic resonance brain images
Expert Syst. Applications
(2018) - et al.
An improved artificial bee colony algorithm based on elite group guidance and combined breadth-depth search strategy
Inform. Sci.
(2018) Multi-channeled MR brain image segmentation: A novel double optimization approach combined with clustering technique for tumour identification and tissue segmentation
Biocybernet. Biomed. Eng.
(2019)- et al.
Brain tumor classification using a hybrid deep autoencoder with Bayesian fuzzy clustering-based segmentation approach
Biocybernet. Biomed. Eng.
(2020)
An expert system for brain tumor detection: Fuzzy C-means with super resolution and convolutional neural network with extreme learning machine
Med. Hypotheses
An efficient brain tumor image segmentation based on deep residual networks (ResNets)
J. King Saud Univ.-Eng. Sci.
Fully automatic grayscale image segmentation based fuzzy C-means with firefly mate algorithm
J. Ambient. Intell. Human. Comput.
Automatic initial boundary generation methods based on edge detectors for the level set function of the Chan-Vese segmentation model and applications in biomedical image processing
“Contrast enhanced medical MRI evaluation using Tsallis entropy and region growing segmentation
J. Ambient Intell. Human. Comput.
Comparative analysis between active contour and otsu thresholding segmentation algorithms in segmenting brain tumor magnetic resonance imaging
J. Inform. Technol. Manag. 12. Spec. Issue: Deep Learn. Visual Inform. Anal. Manag.
Hybridizing Convolutional Neural Network for Classification of Lung Diseases
Int. J. Swarm Intell. Res. (IJSIR)
Brain tumour classification using a hybrid deep autoencoder with Bayesian fuzzy clustering-based segmentation approach
Biocybernet. Biomed. Eng.
A fuzzy based ROI selection for encryption and watermarking in medical image using DWT and SVD
Multimedia Tools Applications
A new approach based on exponential entropy with modified kernel fuzzy c-means clustering for MRI brain segmentation
Evol. Intell.
Brain MR Image Segmentation using a Fast Fuzzy Clustering Approach
2021 12th International Conference on Computing Communication and Networking Technologies (ICCCNT)
“A novel type-2 fuzzy C-means clustering for brain MR image segmentation
IEEE Trans. Cybernet.
BAT algorithm with fuzzy C-ordered means (BAFCOM) clustering segmentation and enhanced capsule networks (ECN) for brain cancer MRI images classification
IEEE Access
Automatic human brain tumor detection in MRI image using template-based K means and improved fuzzy C means clustering algorithm
Big Data Cogn. Comput.
Textural features for image classification
IEEE Trans. Syst. Man Cybernet.
A clinical support system for brain tumor classification using soft computing techniques
J. Med. Syst.
Cited by (6)
The Effective Quantitative Analysis for Brain Tumor Diagnosis Using an Efficient Deep Learning Algorithm
2023, International Journal on Recent and Innovation Trends in Computing and CommunicationTumour region detection in MR brain images using MFCM based segmentation and Self Accommodative JAYA based optimization
2023, ACCESS 2023 - 2023 3rd International Conference on Advances in Computing, Communication, Embedded and Secure SystemsConfidence Region Identification and Contour Detection in MRI Image
2022, Computational Intelligence and NeuroscienceReview on Hybrid Segmentation Methods for Identification of Brain Tumor in MRI
2022, Contrast Media and Molecular Imaging