{"id":62628,"date":"2024-12-30T10:26:26","date_gmt":"2024-12-30T10:26:26","guid":{"rendered":"https:\/\/biomedpharmajournal.org\/?p=62628"},"modified":"2025-01-07T04:07:57","modified_gmt":"2025-01-07T04:07:57","slug":"medical-image-segmentation-using-enhanced-feature-weight-learning-based-fcm-clustering","status":"publish","type":"post","link":"https:\/\/biomedpharmajournal.org\/staging\/vol17no4\/medical-image-segmentation-using-enhanced-feature-weight-learning-based-fcm-clustering\/","title":{"rendered":"Medical Image Segmentation using Enhanced Feature Weight Learning Based FCM Clustering"},"content":{"rendered":"\n

Introduction<\/strong><\/p>\n\n\n\n

Image\nsegmentation is the process of partitioning an image into different parts in\nthe sets of pixels or superpixels. The segmentation is a crucial step for image\nanalysis and its understanding. It aims to represent the image information in a\nform that is more suitable for different applications. Image segmentation finds\na variety of applications such as face recognition1<\/sup>, object\ndetection2<\/sup>, fingerprint recognition3<\/sup>, biomedical image\nprocessing4<\/sup> and industrial applications5-8<\/sup>. Image\nsegmentation may be broadly classified as threshold-based methods9, 10<\/sup>\nregion-extension methods11<\/sup> and clustering based methods12<\/sup>.\nThese methods have their different advantages and limitations. In recent years,\ncluster-based methods have become very popular because of its fast\nimplementation and superior performance13<\/sup>.<\/p>\n\n\n\n

The\nFuzzy c-means (FCM) is a popular data clustering algorithm for colour images due\nto its easy and fast implementation14<\/sup>. In this method, pixels can be\nassigned to different clusters; providing improved information and better\nsegmentation performance15<\/sup> as compared to hard clustering techniques\nsuch as k-means16<\/sup>. In addition, the fuzzy membership set in this method\nhelps discover complex evaluation between samples and clusters with more accuracy.\nThe limitations of FCM algorithm are its sensitivity to the selection of initial\ncluster centres17<\/sup>, and the property that different features of the\nimage contribute equally and hold same importance18, 19<\/sup>. As the number\nof features increase in source images; some features hold more importance than\nothers. So assigning same weights to all features limits the performance of\nimage segmentation20, 21<\/sup>.<\/p>\n\n\n\n

The\nRGB components of an input image are shown in Fig 1. As shown in the figure, R\nchannel should be better than the other channels within the feature of\nsegmentation. This can also be true for other features and sub-features; so assigning\nsame weights reduces the efficiency of image segmentation.<\/p>\n\n\n\n\n\n
\"\"<\/td>\n\n

Figure 1: Color components of RGB image (a) Input image, (b) Component R, (c) Component G, and (d) Component B.<\/strong><\/p>\n

\u00a0<\/p>\n

Click here to view Figure<\/a><\/p>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

\u00a0<\/p>\n\n\n

The paper is organized in the following sections.\nThe current state of research in the area of image segmentation is presented in\nsection 2. The proposed algorithm is presented in section 3. Section 4 describes\nthe results and analysis. Conclusion is presented in section 5.<\/p>\n\n\n\n

Related Work<\/strong><\/p>\n\n\n\n

In recent years, a variety of image segmentation\nalgorithms based on clustering techniques have been proposed.  An algorithm based on combination of modified\nk-means and Imperialistic Competitive Algorithm (ICA) used cluster centers to\nimprove segmentation performance22<\/sup>. \nThe CSFCM method improved the segmentation performance by combining\nthree meta-heuristics of biogeography, genetic and firefly23<\/sup>. A fuzzy\nmodel based on unsupervised learning combined color and Gaussian density into\nfuzzy clustering algorithm and improved clustering by incorporating the\nneighbouring information into the learning step24<\/sup>. A technique named\nA-PSO-IT2IFCM used particle swarm optimization (PSO) to find suitable cluster center\nand the fuzzifiers for image segmentation25<\/sup>. Other similar proposed methods\ntried to obtain required set of clusters or primary centers26-28<\/sup>. G.\nSilva analysed improvements in dental X-ray using automated teeth segmentation29<\/sup>.\nThe article provided insights into current trends and potential future\ndirections in this field by introducing a new dataset and benchmarking different\nsegmentation techniques improve dental diagnostics efficiency and accuracy.<\/p>\n\n\n\n

In the past years, Multi-Objective Evolutionary\nAlgorithms (MOEAs) have been frequently used for color image segmentation25,\n30<\/sup>. The purpose of these algorithms is to\nfind optimal cluster centers. But these methods also suffer from higher\ncomputational requirements. To address this limitation, Kriging-assisted\nreference vector evolutionary algorithm (KREVA) based methods were proposed30,\n31<\/sup>. A new deep\nlearning method combined attention mechanisms and JGate modules into the\nResidual U-Net architecture for brain tumor segmentation32<\/sup>. This\ntechnique successfully extracted contextual information and minute details from\nbrain MRI images to improve segmentation accuracy.<\/p>\n\n\n\n

A\nmethod combined level sets with Backchannel Filling Convolutional Neural\nNetworks (CNNs) for segmentation of skin lesions in images33<\/sup>. Lesion\nboundaries were enhanced by delineation of by filling in missing or unclear\nregions in skin lesion images; leading to improvement in segmentation accuracy.\nA method combined YOLOv8 with SAM and HQ-SAM models to achieve thorough\nmultimodal segmentation in medical imaging34<\/sup>. While SAM and HQ-SAM\nimprove segmentation by utilizing multi-scale features and hierarchical\ncontexts; YOLOv8 offers accurate object detection. The combined method improved\naccuracy and robustness in the segmentation of complex medical images.<\/p>\n\n\n\n

An algorithm named entropy regularized weighted FCM<\/em> (EWFCM) used an improved objective\nfunction using local-feature weighted entropy regularization method, to select\nthe optimal weights for image features35<\/sup>. A clustering method\nproposed an objective function to use local feature weighting scheme for finding\nclusters in images36<\/sup>. A new FCM based cluster weighting and feature\nweighting method overcame limitations in existing methods37<\/sup>. In\nCGGFCM, a feature weighting technique was used to improve clustering accuracy\nand a strategy for automatic cluster weighting to lessen sensitivity to cluster\ninitialization38<\/sup>. <\/p>\n\n\n\n

Early detection and diagnosis of cancer leads\nto successful treatment and lower mortality rates. In recent years, deep\nlearning techniques have been used to achieve improved performance in image\nsegmentation algorithms. An algorithm based on improved fuzzy local information\nC means (IFF-FLICM) segmentation was successfully used for detection and classification\nof Dataset-255 brain tumor images39<\/sup>. A study used a variety of\nmachine learning classification method for predicting cervical cancer using\ndifferent risk factors. Based on this study and analysis of a dataset of around\n850 patients, a deep-learning method was proposed for early diagnosis and\ndetection of cervical cancer40<\/sup>. The flowchart of the proposed method\nincorporates the collection of patient’s information, pre-processing step,\ntraining of model, threshold prediction and setting, validation, and final\ndiagnosis.<\/p>\n\n\n\n

A\nnew machine learning model and FCM based segmentation algorithm was used for\nclassification and detection of breast cancer from mammogram images41<\/sup>.\nFirstly, fuzzy factor improved fast and robust fuzzy c means (FFI-FRFCM)\nsegmentation segmented the input image by modifying the member partition matrix\nof the FRFCM technique. Then an improved particle swarm optimization (PSO) was\nused for weight optimization of the ensemble extreme learning machine (EELM)\nmodel. The proposed method showed improved performance for classification of\nbreast cancer images. Another algorithm for classification of\nbreast cancer images introduced a novel hybrid DenseNet121-based ELM Model42<\/sup>.\nThe features collected after the pooling and flatten layers at the first stage\nof the classification were used as input to the proposed DenseNet121-ELM model.\nThe AdaGrad optimization algorithm was used for updating the weights of ELM. The\nperformance evaluation of proposed method for batch size up to 128 showed improved\nperformance and robustness of algorithm.<\/p>\n\n\n\n

 Many\ntechniques have been proposed for mitigating the limitations of sensitivity to\nequalization and equal importance of features issue in FCM method for image\nsegmentation. In this paper, an extended set of image feature has been used to\nimprove the representation of image information. An improved weight optimization\nmethod has been used to overcome the limitations of FCM; in the quest to\nimprove color image segmentation for medical images.<\/p>\n\n\n\n

Proposed Work<\/strong><\/p>\n\n\n\n

The proposed image segmentation technique is presented\nin Fig. 2. Initially, important features are extracted from the input image; which\nare used in the clustering process. ICA is used for finding the optimal parameters,\nwhich then are used to improve CGFFCM by finding weights of features for\nfeature groups. The proposed algorithm has been described in\nthe following sub-sections.<\/p>\n\n\n\n\n\n
\"\"<\/td>\n\n

Figure 2: Block diagram of the proposed approach<\/strong><\/p>\n

\u00a0<\/p>\n

Click here to view Figure<\/a><\/p>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

\u00a0<\/p>\n\n\n

Feature Extraction <\/strong><\/p>\n\n\n\n

The\nproposed method uses an effective combination of image features, including texture,\nedge, local homogeneity, and CIELAB20, 43<\/sup>.\nThe features represent the important properties of the images and are described\nas follows:<\/p>\n\n\n\n

Local\nhomogeneity:<\/em> In the context of image processing, it refers to the\ndegree of uniformity in pixel values within a small neighbourhood surrounding a\nspecific pixel. It quantifies the extent of constant value of image intensity\nor color in a localized area.\nBy identifying regions with distinct homogeneity values (high for\nuniform areas, low for edges\/textures); the segmentation algorithm groups distinct\npixels into meaningful objects or regions. <\/p>\n\n\n\n\n\n
\"\"<\/td>\n\n

Figure 3: Local homogeneity features (a) Input image (b) Component H (c) Component S (d) Component V<\/strong><\/p>\n

\u00a0<\/p>\n

\u00a0<\/p>\n

Click here to view Figure<\/a><\/p>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

\u00a0<\/p>\n\n\n\n\n
\"\"<\/td>\n\n

Figure 4: Color features from CIELAB colour channels (a) Input image (b) Component L \u00a0(c) Component A (d) Component B<\/strong><\/p>\n

\u00a0<\/p>\n

\u00a0<\/p>\n

Click here to view Figure<\/a><\/p>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

\u00a0<\/p>\n\n\n

Color:<\/em> It is represented in CIELAB pictures (Lab*) by L*, a*, and b* values. The brightness scale Lightness (L)*, goes from 0 (complete black) to 100 (white). \u00a0Value of Green-Red (a)* changes from negative (green) to positive (red). The Yellow-Blue (b)* scale goes from positive (yellow) to negative (blue). <\/p>\n\n\n\n

Grayscale or achromatic colors have values of a* = 0 and b* = 0. Distinct color zones within an image can be identified by analyzing these values for every pixel in the image. <\/p>\n\n\n\n

Texture: <\/em>This feature go beyond\nsimple color gradation to depict recurring patterns in an image by describing\nhow intensity or color varies throughout the image. This is essential for\nsegmenting images, particularly when dividing things according to their surface\ntextures rather than merely their colors. Texture enhances color in\nsegmentation process. Color conveys basic information, while texture describes\nhow color variations are arranged in space. When these features are combined,\nsegmentation accuracy is better than, when color is used alone.<\/p>\n\n\n\n\n\n
\"\"<\/td>\n\n

Figure 5: Texture feature component (a) Input image (b) Output of Gabor filter<\/strong><\/p>\n

\u00a0<\/p>\n

\u00a0<\/p>\n

Click here to view Figure<\/a><\/p>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

\u00a0<\/p>\n\n\n\n\n
\"\"<\/td>\n\n

Figure 6: Edge feature component (a) Input image (b) Edge component<\/strong><\/p>\n

\u00a0<\/p>\n

\u00a0<\/p>\n

Click here to view Figure<\/a><\/p>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

\u00a0<\/p>\n\n\n

Edges:<\/em>\nThe boundaries between objects or regions in an image are crucial for image\nsegmentation. These often arise from significant changes in pixel intensity,\ncolor, or texture. Identifying these edges is the key to separate objects and\nachieve accurate segmentation. Extracting edge features can be done through\nvarious techniques like gradient-based methods. Sobel and Prewitt filters\ncalculate the rate of intensity change between pixels, highlighting areas with\nsharp transitions. The Laplacian operator emphasizes areas with substantial\nintensity variations, with edges corresponding to zero-crossing points.\nFinally, Canny edge detection combines gradient information to locate edges\nprecisely while minimizing false detections. \nWhile edge features offer advantages like simplicity, efficiency, and\nclear interpretability; these can be sensitive to noise and might not capture\nblurry boundaries or provide enough detail for complex textures.<\/p>\n\n\n\n

Clustering Method <\/strong><\/p>\n\n\n\n

The clustering-based algorithms may become sensitive\nto image artifacts if the spatial information is ignored. This will lead to\nshifting of the intensities values of pixels, leading to the situation where pixels\nfrom different clusters may possess similar features. Therefore cluster-based\nalgorithms, look for means to suppress the adverse effect of noise.<\/p>\n\n\n\n

Among\nthe two strategies; one is group-local feature strategy, where weights are allocated\nto the group of features in a cluster. Other is local feature weighting\nstrategy, in this weight is allocated to each feature in the group of feature. Both\nthese methods are used to improve clustering accuracy. An algorithm based on automatic\nweighting of clusters reduced the sensitivity of the segmentation method and\nobtained improved result by using cluster initialization and group feature\nweighting method38<\/sup>. <\/p>\n\n\n\n

In\nimage segmentation, the different group of features can have different\nimportance; so the advantage of group feature weighting is more obvious than local\nweighting which assigns one weight to all the features37<\/sup>. Within a\ngroup also, different features might be less or more important than the other.\nSo assigning same weights to all the components will reduce the quality of\nsegmented image. So, we use features and sub-features with more varying degree of\nimportance in image clusters. The improved weighting of clusters is directly\nrelated to better segmentation performance. Further; to control the sensitivity\nto initial clusters, an effective cluster weighting technique is incorporated.\nThe relative importance of features and sub-features is considered while selecting\nthe weight of image clusters38<\/sup>.<\/p>\n\n\n\n

Optimization <\/strong><\/p>\n\n\n\n

The clustering method is used in combination with the Imperialist Competitive Algorithm (ICA) for optimization of feature weighing step44<\/sup>. It finds optimized values of the weight coefficients in the segmentation method. Imperialist competitive algorithm (ICA) is a classic evolutionary optimization technique which incorporates the general political and social behaviour of imperialist countries in an attempt to dominate weaker countries44<\/sup>. In past few years, various algorithms based on ICA and its improved versions have been proposed; which have successful solved several practical optimization methods. Another advantage of ICA is its higher convergence rate as compared to other evolutionary optimization algorithms; delivering significant results in less time45<\/sup>. The Optimization procedure involving integration of ICA with FCM clustering can be described in following steps:<\/p>\n\n\n\n

Initial Population and Weight Coefficient Ranges<\/strong><\/p>\n\n\n\n

Population Initialization: Generate a random initial population of solutions, where each country represents a weight vector [v1<\/sub>,v2<\/sub>,v3<\/sub>] \u200bfor three feature groups.<\/p>\n\n\n\n

Weight Constraints: Set weight range to [0.1, 0.8] to ensure participation of all groups, preventing any feature from being entirely excluded in initial clustering.<\/p>\n\n\n\n

Objective Function Evaluation<\/strong><\/p>\n\n\n\n

Run FCM clustering for each country and calculate the objective function value based on clustering performance; reflects how well the weights align with optimal feature grouping.<\/p>\n\n\n\n

Imperialist Selection<\/strong><\/p>\n\n\n\n

Selecting\nImperialists:<\/em> Rank\ncountries based on objective function values and select the best-performing\nsolutions as imperialists; the remaining become colonies.<\/p>\n\n\n\n

Empire Configuration:<\/em> Experiment with different numbers of initial imperialists Nimp<\/sub><\/em> = [10,20] to find the best structure for the clustering task.<\/p>\n\n\n\n

Colony Allocation Based on Imperialist Power<\/strong><\/p>\n\n\n\n

Assigning Colonies:<\/em> Allocate colonies to imperialists based on their power, calculated using cost differences and normalized to determine each imperialist\u2019s control proportionally as per eq 1 and eq 2. The values are rounded for the final number of colonies per empire.<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

Assimilation – Moving Colonies Toward Imperialists<\/strong><\/p>\n\n\n\n

Cultural\nAssimilation:<\/em>\nShift colonies incrementally towards their imperialists, with movement\ninfluenced by a random variable \ud835\udc4e and an angle \ud835\udf03 for directional diversity.<\/p>\n\n\n\n

Exploration Parameters:<\/em> Set \u03b2<\/em> = 2 for balanced convergence and \u03b3 = \u03c0\/4 to encourage exploration in multiple directions.<\/p>\n\n\n\n

Update Colony Costs and Improvement Check<\/strong><\/p>\n\n\n\n

Recompute colony costs after movement. If a colony’s cost surpasses its imperialist\u2019s, they swap positions, enhancing the empire with potential improvements.<\/p>\n\n\n\n

Calculate Total Empire Cost<\/strong><\/p>\n\n\n\n

Empire Cost Calculation:<\/em> Compute each empire’s total cost TCn<\/sub><\/em>\u200b, combining the imperialist’s cost and an averaged colony cost.<\/p>\n\n\n\n

Weight Coefficient:<\/em> Set attenuation coefficient \u03be = <\/em>0.1 to balance emphasis on imperialist vs. colony contributions, reinforcing high-quality solutions.<\/p>\n\n\n\n

Inter-Empire Competition for Colonies<\/strong><\/p>\n\n\n\n

Competitive\nRedistribution:<\/em>\nWeaker empires lose colonies to stronger ones based on normalized total costs\nand possession probabilities.<\/p>\n\n\n\n

Roulette Selection:<\/em> A roulette wheel mechanism selects winning empires, promoting dynamic growth of powerful empires and reallocation of resources.<\/p>\n\n\n\n

Empire Elimination<\/strong><\/p>\n\n\n\n

Remove empires with no colonies left, and merge remaining imperialists into stronger empires, preserving potentially viable solutions.<\/p>\n\n\n\n

Termination and Resulting Clusters<\/strong><\/p>\n\n\n\n

The algorithm stops when a single empire remains, marking the most dominant clustering solution.<\/p>\n\n\n\n

Final Clustering Output:<\/em> The remaining empire\u2019s colonies represent the final clusters, with each colony position corresponding to feature weights optimized for FCM clustering. Evaluate clusters with benchmark datasets for segmentation quality.<\/p>\n\n\n\n

Results and Discussion<\/strong><\/p>\n\n\n\n

To\ndemonstrate the performance of proposed segmentation method, a set of medical images\nhave been taken from a variety of environments. The images are taken from Berkley\nimage dataset, The BRATS 2016 skin lesion challenge datasets, including color\nimages having balanced and imbalanced regions. To evaluate the generalizability\nof the proposed method, test images Cystoid\nfluid<\/em> image and Kvasir<\/em> image have\nbeen taken respectively from OPTIMA 13 dataset and Kvasir SEG dataset. Cystoid\nfluid refers to fluid\nthat accumulates within cyst-like spaces, often in the retina or other tissues.\nKvasir is a medical image\ndataset, particularly useful for researchers working on computer-aided\ndiagnosis systems in gastroenterology. The segmentation methods are implemented\nusing MATLAB R2018b, and results of the proposed segmentation method are\ncompared with existing standard methods JGate-AttResUNet32<\/sup>, New\nfuzzy c-means18<\/sup>, Automatic segmenting37<\/sup>, and CGFFCM38<\/sup>.\nThe objectively evaluation of these methods is performed using performance\nmetrics Normalized Mutual Information\n(NMI), Accuracy, and F-score. The performance metrics are described as\nfollows:<\/p>\n\n\n\n

Normalized Mutual Information (NMI)<\/strong><\/p>\n\n\n\n

Instead of focusing on individual pixels, NMI considers both the number of segments (quantity) and their arrangement (spatial distribution) in the image to observe segmentation performance.<\/p>\n\n\n\n

Accuracy<\/strong><\/p>\n\n\n\n

It is defined as the number of clustered pixels divided by the total number of pixels. It measures the degree of similarity between the segmentation result and the given ground truth.<\/p>\n\n\n\n

F-score<\/strong><\/p>\n\n\n\n

This metric calculates the harmonic mean of recall and precision values. This metric is effective in evaluation of imbalanced color regions in the image. <\/p>\n\n\n\n

Subjective Performance Evaluation<\/strong><\/p>\n\n\n\n

For subjective evaluation, the results of proposed segmentation method\nand existing methods are shown in Fig. 7 and Fig 8. The results for images named MRI-1, MRI-2, Teeth, skin lesion,\nbuffalo, <\/em>and bird<\/em> are different\nrows in Fig. 7. The segmentation\nresults of Cystoid fluid<\/em> and Kvasir<\/em>\nimages are shown in Fig 8. It is observed that results\nobtained by proposed method are sharper with clear cut boundaries. It can be\nattributed to enhanced feature set which has a direct impact on the\nsegmentation results. It is observed that proposed algorithm performs better\nthan the existing standard methods.<\/strong><\/p>\n\n\n\n\n\n
\"\"<\/td>\n\n

Figure 7: Segmentation results obtained using the CIELAB space for (a) Input image,(b) JGate-AttResUNet32<\/sup> (c) Automatic segmenting37<\/sup> (d) CGFFCM38<\/sup> (e) Proposed method.<\/strong><\/p>\n

\u00a0<\/p>\n

Click here to view Figure<\/a><\/p>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

\u00a0<\/p>\n\n\n\n\n
\"\"<\/td>\n\n

Figure 8: Segmentation results of Cystoid fluid <\/em>and Kvasir<\/em> images (a) Input image, (b) JGate-AttResUNet32<\/sup> (c) CGFFCM38<\/sup> (d) Proposed method.<\/strong><\/p>\n

\u00a0<\/p>\n

Click here to view Figure<\/a><\/p>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

\u00a0<\/p>\n\n\n

Objective Performance Evaluation<\/strong><\/p>\n\n\n\n

For\nobjective performance evaluation, the results of proposed method are compared\nwith existing clustering-based segmentation techniques JGate-AttResUNet32<\/sup>, New\nfuzzy C-means18<\/sup>, and CGFFCM38<\/sup> using\nmetrics Accuracy, Normalized Mutual Information (NMI),\nand F1-Score. The results for different images are shown in Table 1. <\/p>\n\n\n\n

Table 1: Objective performance evaluation of proposed method<\/strong><\/p>\n\n\n\n\n\n\n\n\n\n\n\n\n
\n

Images<\/strong><\/p>\n<\/td>\n

\n

Metric<\/strong><\/p>\n<\/td>\n

\n

CNN& level sets33<\/sup><\/strong><\/p>\n<\/td>\n

\n

JGate-AttResUNet32<\/sup><\/strong><\/p>\n<\/td>\n

\n

New fuzzy C-means18<\/sup><\/strong><\/p>\n<\/td>\n

\n

CGFFCM38<\/sup><\/strong><\/p>\n<\/td>\n

\n

Proposed method<\/strong><\/p>\n<\/td>\n<\/tr>\n

\n

MRI-1<\/p>\n<\/td>\n

\n

Accuracy<\/p>\n

NMI<\/p>\n

F1-Score<\/p>\n<\/td>\n

\n

83.08<\/p>\n

45.09<\/p>\n

–<\/p>\n<\/td>\n

\n

83.72<\/p>\n

45.29<\/p>\n

71.14<\/p>\n<\/td>\n

\n

95.91<\/p>\n

77.57<\/p>\n

96.90<\/p>\n<\/td>\n

\n

99.59<\/p>\n

88.62<\/p>\n

99.78<\/p>\n<\/td>\n

\n

98.478<\/p>\n

95.524<\/p>\n

93.9212<\/p>\n<\/td>\n<\/tr>\n

\n

MRI-2<\/p>\n<\/td>\n

\n

Accuracy<\/p>\n

NMI<\/p>\n

F1-Score<\/p>\n<\/td>\n

\n

93.51<\/p>\n

76.74<\/p>\n<\/td>\n

\n

93.85<\/p>\n

76.02<\/p>\n

70.16<\/p>\n<\/td>\n

\n

94.65<\/p>\n

78.86<\/p>\n

90.97<\/p>\n<\/td>\n

\n

96.52<\/p>\n

83.21<\/p>\n

95.89<\/p>\n<\/td>\n

\n

96.70<\/p>\n

95.98<\/p>\n

89.42<\/p>\n<\/td>\n<\/tr>\n

\n

Teeth<\/p>\n<\/td>\n

\n

Accuracy<\/p>\n

NMI<\/p>\n

F1-Score<\/p>\n<\/td>\n

\n

98.83<\/p>\n

74.81<\/p>\n

–<\/p>\n<\/td>\n

\n

–<\/p>\n

–<\/p>\n

–<\/p>\n<\/td>\n

\n

99.12<\/p>\n

81.26<\/p>\n

92.92<\/p>\n<\/td>\n

\n

99.46<\/p>\n

87.19<\/p>\n

95.60<\/p>\n<\/td>\n

\n

99.25<\/p>\n

99.56<\/p>\n

92.67<\/p>\n<\/td>\n<\/tr>\n

\n

Skin lesion<\/p>\n<\/td>\n

\n

Accuracy<\/p>\n

NMI<\/p>\n

F1-Score<\/p>\n<\/td>\n

\n

–<\/p>\n

–<\/p>\n

–<\/p>\n<\/td>\n

\n

–<\/p>\n

–<\/p>\n

–<\/p>\n<\/td>\n

\n

73.85<\/p>\n

77.48<\/p>\n

–<\/p>\n<\/td>\n

\n

96.43<\/p>\n

93.77<\/p>\n

86.26<\/p>\n<\/td>\n

\n

96.8<\/p>\n

94.36<\/p>\n

87.60<\/p>\n<\/td>\n<\/tr>\n

\n

Buffalo<\/p>\n<\/td>\n

\n

Accuracy<\/p>\n

NMI<\/p>\n

F1-Score<\/p>\n<\/td>\n

\n

–<\/p>\n

–<\/p>\n

–<\/p>\n<\/td>\n

\n

92.92<\/p>\n

77.02<\/p>\n

83.80<\/p>\n<\/td>\n

\n

98.71<\/p>\n

74.23<\/p>\n

88.63<\/p>\n<\/td>\n

\n

99.68<\/p>\n

91.66<\/p>\n

97.37<\/p>\n<\/td>\n

\n

99.70<\/p>\n

99.858<\/p>\n

97.07<\/p>\n<\/td>\n<\/tr>\n

\n

Bird<\/p>\n<\/td>\n

\n

Accuracy<\/p>\n

NMI<\/p>\n

F1-Score<\/p>\n<\/td>\n

\n

99.18<\/p>\n

81.22<\/p>\n

–<\/p>\n<\/td>\n

\n

99.27<\/p>\n

88.24<\/p>\n

77.96<\/p>\n<\/td>\n

\n

96.12<\/p>\n

31.32<\/p>\n

97.98<\/p>\n<\/td>\n

\n

99.59<\/p>\n

88.62<\/p>\n

99.78<\/p>\n<\/td>\n

\n

99.61<\/p>\n

96.40<\/p>\n

95.68<\/p>\n<\/td>\n<\/tr>\n

\n

Cystoid fluid<\/p>\n<\/td>\n

\n

Accuracy<\/p>\n

NMI<\/p>\n

F1-Score<\/p>\n<\/td>\n

\n

–<\/p>\n

–<\/p>\n

–<\/p>\n<\/td>\n

\n

91.42<\/p>\n

87.35<\/p>\n

76.80<\/p>\n<\/td>\n

\n

–<\/p>\n

–<\/p>\n

–<\/p>\n<\/td>\n

\n

97.34<\/p>\n

93.21<\/p>\n

96.11<\/p>\n<\/td>\n

\n

98.14<\/p>\n

96.15<\/p>\n

95.66<\/p>\n<\/td>\n<\/tr>\n

\n

Kvasir<\/p>\n<\/td>\n

\n

Accuracy<\/p>\n

NMI<\/p>\n

F1-Score<\/p>\n<\/td>\n

\n

–<\/p>\n

–<\/p>\n

–<\/p>\n<\/td>\n

\n

92.22<\/p>\n

87.92<\/p>\n

78.46<\/p>\n<\/td>\n

\n

–<\/p>\n

–<\/p>\n

–<\/p>\n<\/td>\n

\n

97.81<\/p>\n

89.71<\/p>\n

98.08<\/p>\n<\/td>\n

\n

97.13<\/p>\n

95.32<\/p>\n

96.22<\/p>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n\n\n

From Table 1, it can be observed that proposed\nmethod provides best value of most of the metrics. The superior performance of\nproposed method can be attributed to enhanced feature set and use of optimal weights\nof groups and sub features in the proposed group feature weighting method. So\nit be concluded that for medical images, the proposed method delivers improved\nsegmentation performance as compared to existing techniques. <\/p>\n\n\n\n

The propose algorithm provides improved\nperformance due to improved feature extraction and optimizations steps. Computational\ntime of proposed segmentation is compared with other segmentation methods in\nTable 2. It is observed that computational time of proposed algorithm is around\n15% higher than CGFFCM. As the processing capabilities are improving, the\nhigher computational cost can be accepted for improved performance for most of\nthe segmentation applications, although this can limit the real-time\napplications of proposed algorithm.<\/p>\n\n\n\n

Table 2: Average Computational time (seconds) of segmentation methods<\/strong><\/p>\n\n\n\n\n\n
\n

CNN& level sets33<\/sup><\/strong><\/p>\n<\/td>\n

\n

JGate-AttResUNet32<\/sup><\/strong><\/p>\n<\/td>\n

\n

New fuzzy C-means18<\/sup><\/strong><\/p>\n<\/td>\n

\n

CGFFCM38<\/sup><\/strong><\/p>\n<\/td>\n

\n

Proposed method<\/strong><\/p>\n<\/td>\n<\/tr>\n

\n

25.23<\/p>\n<\/td>\n

\n

29.19<\/p>\n<\/td>\n

\n

24.67<\/p>\n<\/td>\n

\n

24.93<\/p>\n<\/td>\n

\n

28.71<\/p>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n\n\n

Further,\nlike other FCM-based methods, the proposed method\u2019s dependence on local feature\ninformation and pixel intensities can make it difficult to maintain performance\nfor noisy images; as noise can distort the clustering process and lead to\ninaccurate segmentation. These effects can be controlled to some extent by\npre-processing the source images before segmentation process.<\/p>\n\n\n\n

Conclusion<\/strong><\/p>\n\n\n\n

In\nthis paper, a medical segmentation technique based on group feature weighting and\ncluster weighing scheme is proposed. Feature extraction is done using Homogeneity,\nCIELAB, texture and edge. The\nclustering step is used with Imperialist Competitive Algorithm for feature\nweight optimization. The segmentation performance of the proposed\nmethod is compared with existing algorithm in subjective and objective evaluations\non medical images using a variety of parameters. Through performance\nevaluation, it is established that proposed algorithm provides superior\nsegmentation performance as compared with existing techniques.<\/p>\n\n\n\n

Acknowledgement<\/strong><\/p>\n\n\n\n

The authors would like to thank Department of ECE, University Institute of Engineering & Technology, Panjab University, Chandigarh for providing guidance and laboratory facilities for carrying out the research work.<\/p>\n\n\n\n

Funding Sources<\/strong> <\/p>\n\n\n\n

The author(s) received no financial support for the research, authorship, and\/or publication of this article <\/p>\n\n\n\n

Conflict of Interest<\/strong><\/p>\n\n\n\n

Authors have no conflict of interest to declare.<\/p>\n\n\n\n

Conflict of Interest<\/strong><\/p>\n\n\n\n

The authors do not have any conflict of\ninterest<\/p>\n\n\n\n

Data Availability Statement<\/strong><\/p>\n\n\n\n

This statement does\nnot apply to this article<\/p>\n\n\n\n

Ethics Statement<\/strong><\/p>\n\n\n\n

This research did not involve human participants, animal subjects, or\nany material that requires ethical approval <\/strong><\/strong><\/p>\n\n\n\n

Informed Consent Statement<\/strong><\/p>\n\n\n\n

This study did not involve human participants, and therefore, informed consent was not required <\/p>\n\n\n\n

Clinical Trial Registration<\/strong> <\/p>\n\n\n\n

This research does not involve any clinical trials <\/p>\n\n\n\n

Authors Contribution<\/strong><\/p>\n\n\n\n

Ashima Koundal contributed in conceptualising the idea, implementing\nthe algorithm, and writing the manuscript. <\/p>\n\n\n\n

Dr.\nSumit Budhiraja contributed in designing the analysis, implementing the\nalgorithm, analysis of results, reviewing and editing the manuscript.<\/p>\n\n\n\n

Dr. Sunil Agrawal contributed in analysis and validation of results. Each author mentioned has significantly and directly contributed intellectually to the project and has given their approval for its publication.<\/p>\n\n\n\n

References\n<\/strong><\/p>\n\n\n\n

  1. Farajzadeh N, Hashemzadeh M. Exemplar-based facial expression recognition. Information Sciences<\/em>. 2018;460-461:318-330.
    CrossRef<\/a><\/li>
  2. Farajzadeh N, Karamiani A, Hashemzadeh M. A fast and accurate moving object tracker in active camera model. Multimed Tools Appl.<\/em>, 2018;77(6): 6775-6797.
    CrossRef <\/a><\/li>
  3. Sharma R P, Dey S. Two-stage quality adaptive fingerprint image enhancement using Fuzzy C-means clustering based fingerprint quality analysis. Image Vis Comput.<\/em> 2019;83\u201384:1\u201316.
    CrossRef <\/a><\/li>
  4. Hashemzadeh M, Azar B A. Retinal blood vessel extraction employing effective image features and combination of supervised and unsupervised machine learning methods. Artif Intell Med<\/em>. 2019;95:1-15.
    CrossRef <\/a><\/li>
  5. Hashemzadeh M, Farajzadeh N. A Machine Vision System for Detecting Fertile Eggs in the Incubation Industry. International Journal of Computational Intelligence Systems<\/em>. 2016;9: 850-862.
    CrossRef <\/a><\/li>
  6. Hashemzadeh M. Hiding information in videos using motion clues of feature points. Computers & Electrical Engineering<\/em>. 2018;68:14-25.
    CrossRef <\/a><\/li>
  7. Hashemzadeh M, Asheghi B, Farajzadeh N. Content-aware image resizing: An improved and shadow-preserving seam carving method. Signal Processing<\/em>. 2019;155:233-246.
    CrossRef <\/a><\/li>
  8. Hashemzadeh M, Zademehdi A. Fire detection for video surveillance applications using ICA K-medoids-based color model and efficient spatio-temporal visual features. Expert Syst Appl<\/em>. 2019; 130:60-78.
    CrossRef <\/a><\/li>
  9. Lifang H, Songwei H. Modified firefly algorithm based multilevel thresholding for color image segmentation. Neurocomputing<\/em>. 2017; 240: 152-174.<\/em>
     CrossRef <\/a><\/li>
  10. Pare S, Kumar A, Bajaj V, Singh G K. A multilevel color image segmentation technique based on cuckoo search algorithm and energy curve. Appl Soft Comput<\/em>. 2016;47:76-102.
    CrossRef <\/a><\/li>
  11. Feng L, Li H, Gao Y, Zhang Y. A Color Image Segmentation Method Based on Region Salient Color and Fuzzy C-Means Algorithm. Circuits Syst Signal Process<\/em>. 2020;39:586-610.
    CrossRef <\/a><\/li>
  12. Tan K S, Isa N A M, Lim W H. Color image segmentation using adaptive unsupervised clustering approach. Appl Soft Comput<\/em>. 2013;13(4):2017-2036.
    CrossRef <\/a><\/li>
  13. Farshi T R, Drake J H, \u00d6zcan E. A multimodal particle swarm optimization-based approach for image segmentation. Expert Syst Appl<\/em>. 2020;149:113233.
    CrossRef <\/a><\/li>
  14. Son L H, Tuan T M. Dental segmentation from X-ray images using semi-supervised fuzzy clustering with spatial constraints. Eng Appl Artif Intell<\/em>. 2017;59:186-195.
    CrossRef <\/a><\/li>
  15. Zhang X, Jian M, Sun Y, Wang H, Zhang C. Improving image segmentation based on patch-weighted distance and fuzzy clustering. Multimedia Tools and Applications<\/em>. 2020;79:633-657.
    CrossRef <\/a><\/li>
  16. MacQueen J. Some methods for classification and analysis of multivariate observations. Proceedings of the 5th Berkeley Symposium on Mathematical Statistics and Probability<\/em>. 1967;1: 281\u2212297. <\/li>
  17. Choy SK, Yuen K, Yu C. Fuzzy bit-plane-dependence image segmentation R. Signal Processing<\/em>. 2019;154:30\u201344.
    CrossRef <\/a> <\/li>
  18. Hashemzadeh M, Oskouei A G, Farajzadeh N. New fuzzy C-means clustering method based on feature-weight and cluster-weight learning. Appl Soft Comput<\/em>. 2019;78:324-45.
    CrossRef <\/a> <\/li>
  19. Pimentel BA, de Souza RMCR. Multivariate Fuzzy C-Means algorithms with weighting. Neurocomputing<\/em>. 2016;174B:946-65.
    CrossRef <\/a> <\/li>
  20. Zhou Z, Zhao X, Zhu S. K-harmonic means clustering algorithm using feature weighting for color image segmentation. Multimedia Tools and Applications <\/em>2018;77:15139-15160.
    CrossRef <\/a> <\/li>
  21. Xing HJ, Ha MH. Further improvements in Feature-Weighted Fuzzy C-Means. Information Sciences<\/em>. 2014; 267:1-15.
    CrossRef <\/a><\/li>
  22. Babrdelbonab M, Mohd Hashim SZ, Bazin NEN. Data analysis by combining the modified k-means and imperialist competitive algorithm. J Teknol<\/em>. 2014;70(5):51-57.
    CrossRef <\/a><\/li>
  23. Abdellahoum H, Mokhtari N, Brahimi A, Boukra A. CSFCM: An improved fuzzy C-Means image segmentation algorithm using a cooperative approach. Expert Syst Appl<\/em>. 2021;166:114063.
    CrossRef <\/a><\/li>
  24. Choy SK, Ng C, Yu C. Unsupervised fuzzy model-based image segmentation. Signal Processing<\/em>. 2020;171:107483.
    CrossRef <\/a><\/li>
  25. Zhao F, Chen Y, Liu H, Fan J. Alternate PSO-Based Adaptive Interval Type-2 Intuitionistic Fuzzy C-Means Clustering Algorithm for Color Image Segmentation. IEEE Access<\/em>. 2019;7:64028-64039.
    CrossRef <\/a><\/li>
  26. Mikaeil R, Haghshenas SS, Haghshenas SS, Ataei M. Performance prediction of circular saw machine using imperialist competitive algorithm and fuzzy clustering technique. Neural Comput Appl.<\/em> 2018;29:283-292.
    CrossRef <\/a><\/li>
  27. Aliniya Z, Mirroshandel SA. A novel combinatorial merge-split approach for automatic clustering using imperialist competitive algorithm. Expert Syst Appl.<\/em> 2019;117:243-266.
    CrossRef <\/a><\/li>
  28. Niknam T, Fard ET, Ehrampoosh S, Rousta A. A new hybrid imperialist competitive algorithm on data clustering. Sadhana – Academy Proceedings in Engineering Sciences<\/em>. 2011; 36:293\u2013315.
    CrossRef <\/a><\/li>
  29. Silva G, Oliveira L, Pithon M. Automatic segmenting teeth in X-ray images: Trends, a novel data set, benchmarking and future perspectives. Expert Systems with Applications<\/em>. 2018;107:15\u201331.
    CrossRef <\/a><\/li>
  30. Zhao F, Zeng Z, Liu H, Lan R, Fan J. Semisupervised Approach to Surrogate-Assisted Multiobjective Kernel Intuitionistic Fuzzy Clustering Algorithm for Color Image Segmentation. IEEE Transactions on Fuzzy Systems<\/em>. 2020;28(6):1023-1034.
    CrossRef <\/a><\/li>
  31. Chugh T, Jin Y, Miettinen K, Hakanen J, Sindhya K. A Surrogate-Assisted Reference Vector Guided Evolutionary Algorithm for Computationally Expensive Many-Objective Optimization. IEEE Transactions on Evolutionary Computation<\/em>. 2018;22(1):129-142.
    CrossRef <\/a><\/li>
  32. Ruba T, Tamilselvi R, Beham MP. Brain tumor segmentation using JGate-AttResUNet \u2013 A novel deep learning approach. Biomed Signal Process Control<\/em>. 2023;84: 104926.
    CrossRef <\/a><\/li>
  33. Huang L, Zhao YG, Yang TJ. Skin lesion image segmentation by using backchannel filling CNN and level sets. Biomed Signal Process and Control<\/em>. 2024;87A:105417.
    CrossRef <\/a><\/li>
  34. Pandey S, Chen KF, Dam EB. Comprehensive Multimodal Segmentation in Medical Imaging: Combining YOLOv8 with SAM and HQ-SAM Models. IEEE\/CVF International Conference on Computer Vision Workshops (ICCVW 2023)<\/em>. 2023;2584-2590.
    CrossRef <\/a><\/li>
  35. Zhou J, Chen L, Chen CLP, Zhang Y, Li HX. Fuzzy clustering with the entropy of attribute weights. Neurocomputing<\/em>. 2016;198:125-134.
    CrossRef <\/a><\/li>
  36. Zhou Z, Zhu S. Kernel-based multiobjective clustering algorithm with automatic attribute weighting. Soft comput<\/em>. 2018;22:3685-3709.
    CrossRef <\/a><\/li>
  37. Hashemzadeh M, Oskouei AG, Farajzadeh N. New fuzzy C-means clustering method based on feature-weight and cluster-weight learning. Applied Soft Computing Journal<\/em>. 2019;78:324-345.
    CrossRef <\/a><\/li>
  38. Oskouei GA, Hashemzadeh M. CGFFCM: A color image segmentation method based on cluster-weight and feature-weight learning. Software Impacts<\/em>. 2022;11:100228.
    CrossRef <\/a><\/li>
  39. Dash S, Siddique M, Mishra S, Gelmecha DJ, Satapathy,S, Rathee DS, Singh RS. Brain Tumor Detection and Classification Using IFF-FLICM Segmentation and Optimized ELM Model. Journal of Engineering, 2024, Article ID 8419540, 2024:1-24.
    CrossRef <\/a> <\/li>
  40. Mohapatra S, Siddique M, Paikaray BK, Riyazbanu S, Automated Invasive Cervical Cancer Disease Detection at Early Stage Through Deep Learning, International Journal of Bioinformatics Research and Applications<\/em>, 2023;19(4):306-326.
     CrossRef <\/a> <\/li>
  41. Pattnaik RK, Siddique M, Mishra S, Gelmecha DJ, Singh RS, Satapathy S. Breast cancer detection and classification using metaheuristic optimized ensemble extreme learning machine. International Journal of Information Technology<\/em>, 2023;15(8):4551-4563.
    CrossRef <\/a> <\/li>
  42. Pattanaik RK, Mishra S, Siddique M, Gopikrishna T, Satapathy S. Breast Cancer Classification from Mammogram Images Using Extreme Learning Machine-Based DenseNet121 Model. Journal of Sensors<\/em>. 2022, Article ID 2731364, 2022:1-12.
    CrossRef <\/a> <\/li>
  43. Gamino-S\u00e1nchez F, Hern\u00e1ndez-Guti\u00e9rrez I V, Rosales-Silva AJ, Gallegos-Funes FJ, M\u00fajica-Vargas D, Ramos-D\u00edaz E, Carvajal-G\u00e1mez BE, Kinani JMV. Block-Matching Fuzzy C-Means clustering algorithm for segmentation of color images degraded with Gaussian noise. Engineering Applications of Artificial Intelligence<\/em>. 2018;73:31\u201349.
    CrossRef <\/a> <\/li>
  44. Talatahari S, Farahmand Azar B, Sheikholeslami R, Gandomi AH. Imperialist competitive algorithm combined with chaos for global optimization. Commun in Nonlinear Sci Numer Simul<\/em>. 2012;17(3):1312-1319.
    CrossRef <\/a> <\/li>
  45. Peri D. Hybridization of the imperialist competitive algorithm and local search with application to ship design optimization. Computers and Industrial Engineering<\/em>. 2019;137:106069.
    CrossRef <\/a><\/li><\/ol>\n","protected":false},"excerpt":{"rendered":"

    Introduction Image segmentation is the process of partitioning an image […]<\/p>\n","protected":false},"author":15,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[119],"tags":[],"class_list":["post-62628","post","type-post","status-publish","format-standard","hentry","category-vol17no4"],"_links":{"self":[{"href":"https:\/\/biomedpharmajournal.org\/staging\/wp-json\/wp\/v2\/posts\/62628","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/biomedpharmajournal.org\/staging\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/biomedpharmajournal.org\/staging\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/biomedpharmajournal.org\/staging\/wp-json\/wp\/v2\/users\/15"}],"replies":[{"embeddable":true,"href":"https:\/\/biomedpharmajournal.org\/staging\/wp-json\/wp\/v2\/comments?post=62628"}],"version-history":[{"count":5,"href":"https:\/\/biomedpharmajournal.org\/staging\/wp-json\/wp\/v2\/posts\/62628\/revisions"}],"predecessor-version":[{"id":63559,"href":"https:\/\/biomedpharmajournal.org\/staging\/wp-json\/wp\/v2\/posts\/62628\/revisions\/63559"}],"wp:attachment":[{"href":"https:\/\/biomedpharmajournal.org\/staging\/wp-json\/wp\/v2\/media?parent=62628"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/biomedpharmajournal.org\/staging\/wp-json\/wp\/v2\/categories?post=62628"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/biomedpharmajournal.org\/staging\/wp-json\/wp\/v2\/tags?post=62628"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}