Malignant melanoma diagnosis applying a machine learning method based on the combination of nonlinear and texture features
Graphical abstract
Introduction
Melanoma can be considered the most dangerous form of skin cancer, composed the most significant portion of the death corresponding to skin cancer [1]. If these malignant cancerous cells can be diagnosed in the primary stage, the chance of saving lives will be increased considerably [2]. Hence, the precise differentiating between the melanoma and the other pigmented skin lesions (non-melanoma) images has remained a severe challenge for dermatologists [3], [4]. The most common diagnostic method physicians employ to detect skin lesion types is the “ABCD” technique. By measuring four morphological specifications, skin lesions to the Non-melanoma or malignant categories can be detected [5], [6]. Asymmetry, border irregularities, color distributions, and diameter are the mentioned features that composed this method. However, due to the drawbacks of the “ABCD” technique in detecting small or primary-stage melanoma in which the irregularities in its boundary are not composed, this method's accuracy is not significant [7], [8]. It implies a challenging task even for dermatologists to diagnose skin cancer due to the different appearance of skin lesions by this noninvasive experimental technique [9].
On the other hand, computer-aided systems for diagnosing skin lesions are a growing area of research, created a lot of interest in developing CAD systems through both machine learning and deep learning methods [4], [10]. The CAD system can be considered a 'second opinion' to help radiologists and dermatologists decide [11], [12]. Decreasing the workload, reducing the false-negative diagnosis due to probabilistic physician mistakes, and avoiding overloaded ignoring are the main advantages of CAD systems [13], [14]. These methods usually involve three significant steps:
- i.
Skin boundary detection, ii. Feature extraction, iii. Classification
The boundary detection process segments the skin lesion images and extracts their ROI, which is critical for the precise classification of skin lesions [15]. The feature extraction process uses visual properties such as lesion shape, and texture information. Finally, the classifiers are employed to determine the new samples are belonging to which categories.
Due to melanoma diagnosis at an early stage has a vital role in its treatment, several methods are developed in recent years for melanoma detection. In this section, some of these approaches have been discussed. In [15], the BEMD technique has been utilized for segmentation and enhancing dermoscopic images employing some benchmark databases.
Cheong et al extracted entropy and energy values of the skin lesion dataset and classified them by SVM and Radial Basis Function (RBF), obtaining the best accuracy result of 97.50 %. Zakeri et al used Dr. H. Ganster of the department of dermatology as the database and applied the median filter and SVD transform for the segmentation stage. Besides, Grey Level Co-occurrence Matrix (GLCM) is extracted as feature, and the hybrid classifier is designed as a classifier. The accuracies of 96.8 %, 97.3 %, and 98.8 % for melanoma, dysplastic, and Non-melanoma lesions are achieved, respectively [16]. In [17], Afed et al extracted the Region Of Interest (ROI) of images by the Statistical Region Merging (SRM) algorithm and combines different textural and color features. Furthermore, they used several classifiers including SVM, AdaBoost, and Artificial Neural Network (ANN) to acquire 92.96 % accuracy, 84.78 % specificity, and 96.04 % sensitivity. ACS and ISIC images have been used as databases in [18], and mathematical operations for segmentation. Geometric, texture, and statistical features are extracted for these images, and SVM is used as a classifier. Dezhong et al utilized the CWCO algorithm or features selection and the best correct detection rate is- 92.64 % for ACS images and 87.5 % for ISIC images. Applying some related internet sites, skin data set images are provided in [19]. Dorj et al used ECOC SVM, and deep convolutional neural network as classifiers, and 95.1 % accuracy, 94.17 % specificity, and 98.9 % sensitivity are attained as the result. Madooei et al in [20] used PH2 dermoscopic images and Mean-shift × tool EDISON for the dataset and segmentation approach, respectively. Besides, Blue-Whitish Structure (BWS) features are extracted and Multiple Instance Learning (MIL) is utilized as a classifier. In this work, the accuracy and specificity are 84.50 % and 87.90 %. Xanthous race and a combination of SGNN with manual interaction are the dataset and segmentation stage of [21]. Introducing the novel border features like color, texture, and shape for feature extraction, and Neural Network Meta Ensemble Model and PCA for classification and feature selection, the evaluation results for this work are accuracy: 94.17 %, specificity: 93.75 % sensitivity: 95 %. [22] present a fully automated method for segmenting the skin melanoma at its earliest stage by employing a deep-learning-based approach, namely faster region-based convolutional neural networks (RCNN) along with fuzzy k-means clustering (FKM) to segment the melanoma-affected portion of skin with variable size and boundaries. Also, a network ensemble classifier is designed that combines back propagation (BP) neural networks with fuzzy neural networks to improve performance. The Average accuracy of 95.4 %, 93.1 %, and 95.6 % on the ISIC-2016, ISIC-2017, and PH2 datasets are achieved, respectively. Melanoma diagnosis based on dermoscopic images is also considered in [23], [24], [26]. In [23], shape, color, and GLRLM features are extracted and KNN, SVM, and DT are used for classification. Besides, PSO-based feature optimization is utilized for feature selection resulting in 95.23 % (10-Fold) and 96.45 % (Hold-Out) evaluation indices. In [24], Ain et al extracted LBP and the other non-image domain-specific features for the skin lesion images and then used KNN, NB, SVM, J48, RF, and MLP for classification. In addition, the GP method is utilized as feature selection leads to an accuracy of 95.83 %. Using ISIC for the dataset and dynamic graph cut algorithm for segmentation in [25] caused an accuracy of 94.3 % for Non-melanoma cases, 91.2 % for melanoma, and 92.9 % for keratosis. In this work, GLCM and color features are extracted from the dataset images and put into the Naïve Bayes classifier. Nasir et al used the fusion of active contour-based for segmentation and extracted color, texture, and H.O.G features. Additionally, SVM is applied as the classifier and the evaluation indices of this work are calculated as accuracy: 97.5 %, specificity: 96.7 %, and sensitivity: 97.7 % [26].
The significant contributions of this work can be considered as follows:
- •
Using ORACM as a new fast and accurate segmentation approach for skin lesion segmentation. As we know, the application of this method for dermoscopic images is not reported in the literature.
- •
Compared to the related literature in which the nonlinear analysis of skin images has gotten less attention, the various types of the more important complexity measures, i.e., Fractal Dimension (FD), Lyapunov Exponent (LE), and entropy, are applied here. These features can obtain the different aspects of the chaotic nature of cancerous tumors related to asymmetry and border irregularity.
- •
GLCM as texture features are utilized to attain the information inside the image and calculate the skin lesion characteristics. Contrast, correlation, energy, homogeneity, and diameter represent the coarseness, linear dependency, textural uniformity, and pixel distribution of the texture, respectively. Furthermore, a combination of nonlinear and texture features is introduced in this paper to exhibit the various aspects of skin lesions.
- •
Applying a metaheuristic multi-objective optimization approach, NSGA II, to reduce the objective function and the selected feature's number simultaneously.
- •
Utilizing k-fold cross-validation to reduce the sensitivity of classification accuracy to the training and testing datasets as well as employ diverse machine learning methods for dermoscopic image classification.
The organization of the rest of this work is given in the following. The description of the dermoscopic images dataset and various steps of image segmentation, feature extraction, feature selection, classification techniques, and classification function are given in Section 2. Section 3 contains the experimental results as well as the performance evaluation. While the proposed method is compared with other similar works in Section 4, the discussion about the obtained results is outlined in Section 5. Finally, this paper is concluded in Section 6. This study's proposed method for classifying the skin lesions and diagnosis the melanoma using dermoscopic images based on the above-mentioned stages is described in detail in Fig. 1.
Section snippets
Method
In this section, the proposed method to diagnose malignant melanoma from dermoscopic images is realized. The main parts of this section are describing the image database, pre-processing feature extraction, feature selection, classification techniques, and classifier evaluation terms.
Experimental results
The proposed approach for type detection of skin lesions is implemented on dermatoscopic images. All methods are simulated on a 3.6 GHz Core i7-4720 CPU system employing MATLAB R2018a (Math works Inc.). The first phase of simulation is devoted to assessing the ability of the ORACM in segmenting and extracting the ROI of the skin lesion, as illustrated in Fig. 4.
As can be seen, this approach detects the ROI of lesions precisely. In the second phase, the introduced nonlinear indices are measured
Comparison with existing schemes:
Some experiments are performed to assess the proposed method's efficiency as the combination of texture and nonlinear indices in skin cancer diagnosis. Selected complexity measures and texture features are employed separately for the classification stages, and the results are presented in Table 7, Table 8 for the nonlinear and texture features, respectively.
Moreover, the accuracy of five designed classifiers using three feature sets consisting of nonlinear, texture, and combination are compared
Discussion
The first section of the experiments is dedicated to segmenting the skin lesion ROI. As can be observed, employing the ORACM method leads to appropriate precision in image segmentation. Eliminating two significant shortcomings of ACM consists of the slow speed of algorithm convergency as well as the sensitivity of the results to the parameters tuning contributes to the ORACM being employed as a fast and low computational cost method for image segmentation.
Besides, Table 1 demonstrates that due
Conclusion
In the present research, a multi-objective optimization algorithm selects a combination of nonlinear indices and texture features to distinguish the skin lesion types. Disorder growth of cancerous cells originated from the chaotic essence of its causation process, causing the complex measures can reflect the asymmetry and border irregularities of skin lesions. On the other hand, the texture features attained from LH and HL sub-bands represent the image content's information. The selection of
CRediT authorship contribution statement
Sepehr Salem Ghahfarrokhi: Software, Writing – original draft. Hamed Khodadadi: Conceptualization, Validation, Supervision, Project administration. Hamid Ghadiri: Methodology, Formal analysis. Fariba Fattahi: Writing – review & editing.
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 (47)
- et al.
A fractal analysis of skin pigmented lesions using the novel tool of the variogram technique
Chaos, Solitons Fractals
(2006) - et al.
Computerized analysis of pigmented skin lesions: a review
Artif. Intell. Med.
(2012) Artificial intelligence-based image classification methods for diagnosis of skin cancer: Challenges and opportunities
Comput. Biol. Med.
(2020)Incorporating clinical knowledge with constrained classifier chain into a multimodal deep network for melanoma detection
Comput. Biol. Med.
(2021)- et al.
A novel melanoma prediction model for imbalanced data using optimized SqueezeNet by bald eagle search optimization
Comput. Biol. Med.
(2021) Boundary-aware context neural network for medical image segmentation
Med. Image Anal.
(2022)Finding reduced Raman spectroscopy fingerprint of skin samples for melanoma diagnosis through machine learning
Artif. Intell. Med.
(2021)A survey on deep learning in medical image analysis
Med. Image Anal.
(2017)Artificial intelligence in screening mammography: a population survey of women’s preferences
Journal of the American College of Radiology
(2021)“An automated skin melanoma detection system with melanoma-index based on entropy features.” Biocybernetics and Biomedical
Engineering
(2021)
Improvement in the diagnosis of melanoma and dysplastic lesions by introducing ABCD-PDT features and a hybrid classifier
Biocybernetics and Biomedical Engineering
Bagged textural and color features for melanoma skin cancer detection in dermoscopic and standard images
Expert Syst. Appl.
Computer-aided skin cancer diagnosis based on a New meta-heuristic algorithm combined with support vector method
Biomed. Signal Process. Control
Intelligent skin cancer detection using enhanced particle swarm optimization
Knowl.-Based Syst.
Skin disease detection and segmentation using dynamic graph cut algorithm and classification through Naive Bayes classifier
Measurement
ORACM: Online region-based active contour model
Expert Syst. Appl.
Analysis of the contour structural irregularity of skin lesions using wavelet decomposition
Pattern Recogn.
Fractal dimension of India using multicore parallel processing
Computers Geosciences
Heart arrhythmia diagnosis based on the combination of morphological, frequency and nonlinear features of ECG signals and metaheuristic feature selection algorithm
Expert Syst. Appl.
Multiscale permutation entropy for two-dimensional patterns
Pattern Recogn. Lett.
Detection of melanoma in dermoscopic images by integrating features extracted using handcrafted and deep learning models
Comput. Ind. Eng.
Water wave optimization: a new nature-inspired metaheuristic
Computers Operations Research
Designing a retrieval-based diagnostic aid using effective features to classify skin Lesion in dermoscopic images
Procedia Comput. Sci.
Cited by (8)
Exosomal microRNAs in regulation of tumor cells resistance to apoptosis
2024, Biochemistry and Biophysics ReportsAdvancements in traditional machine learning techniques for detection and diagnosis of fatal cancer types: Comprehensive review of biomedical imaging datasets
2024, Measurement: Journal of the International Measurement ConfederationLearning how to detect: A deep reinforcement learning method for whole-slide melanoma histopathology images
2023, Computerized Medical Imaging and GraphicsA systematic literature survey on skin disease detection and classification using machine learning and deep learning
2024, Multimedia Tools and ApplicationsA new proposed GLCM texture feature: modified Rényi Deng entropy
2023, Journal of Supercomputing