Malignant melanoma diagnosis applying a machine learning method based on the combination of nonlinear and texture features

https://doi.org/10.1016/j.bspc.2022.104300Get rights and content

Highlights

Abstract

Skin cancer affects people of all skin tones, including those with darker complexions. Melanomas are known as malignant tumors of skin cancer, resulting in an adverse prognosis, responsible for most deaths relating to skin cancer. Early diagnosis and treatment of skin cancer from dermoscopic images can significantly reduce mortality and save lives. In this paper, Computer-Aided Diagnosis (CAD) system based on machine learning algorithms is provided to classify various skin cancer types. The proposed method uses the Online Region-based Active Contour Model (ORACM) with a new binary level set equation and regularization operation to extract skin lesions' Region of Interest (ROI). Additionally, various combinations of different textures and nonlinear features are extracted for the ROI to show the multiple aspects of skin lesions. Several metaheuristic optimization algorithms are used to remove redundant or irrelevant features and reduce the feature space dimension. These are applied to the combination of the extracted features in which, the Non-dominated Sorting Genetic Algorithm (NSGA II) as a multi-objective optimization algorithm has the best performance. Furthermore, various machine learning algorithms including K-Nearest Neighbor (KNN), Support Vector Machine (SVM), Fitting neural network (Fit net), Feed-Forward neural network (FF net), and Pattern recognition network (Pat net) are employed for the classification. In this paper, the best-obtained evaluation parameters based on fivefold cross-validation belong to the combination of selected nonlinear indices, texture, and GLCM features with an accuracy: 92.24 %, specificity: 100 %, and sensitivity: 100 %, resulting through NSGA II and applying the pattern net classifier. Besides, the comparison between this paper's experimental results and other similar works demonstrates the proposed method's efficiency.

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)

  • A. Zakeri et al.

    Improvement in the diagnosis of melanoma and dysplastic lesions by introducing ABCD-PDT features and a hybrid classifier

    Biocybernetics and Biomedical Engineering

    (2018)
  • N. Alfed et al.

    Bagged textural and color features for melanoma skin cancer detection in dermoscopic and standard images

    Expert Syst. Appl.

    (2017)
  • D. Bi

    Computer-aided skin cancer diagnosis based on a New meta-heuristic algorithm combined with support vector method

    Biomed. Signal Process. Control

    (2021)
  • T.Y. Tan

    Intelligent skin cancer detection using enhanced particle swarm optimization

    Knowl.-Based Syst.

    (2018)
  • V. Balaji

    Skin disease detection and segmentation using dynamic graph cut algorithm and classification through Naive Bayes classifier

    Measurement

    (2020)
  • M.F. Talu

    ORACM: Online region-based active contour model

    Expert Syst. Appl.

    (2013)
  • L. Ma et al.

    Analysis of the contour structural irregularity of skin lesions using wavelet decomposition

    Pattern Recogn.

    (2013)
  • A. Husain

    Fractal dimension of India using multicore parallel processing

    Computers Geosciences

    (2022)
  • V. Mazaheri et al.

    Heart arrhythmia diagnosis based on the combination of morphological, frequency and nonlinear features of ECG signals and metaheuristic feature selection algorithm

    Expert Syst. Appl.

    (2020)
  • C. Morel et al.

    Multiscale permutation entropy for two-dimensional patterns

    Pattern Recogn. Lett.

    (2021)
  • P. Bansal et al.

    Detection of melanoma in dermoscopic images by integrating features extracted using handcrafted and deep learning models

    Comput. Ind. Eng.

    (2022)
  • Y.-J. Zheng

    Water wave optimization: a new nature-inspired metaheuristic

    Computers Operations Research

    (2015)
  • L. Singh et al.

    Designing a retrieval-based diagnostic aid using effective features to classify skin Lesion in dermoscopic images

    Procedia Comput. Sci.

    (2020)
  • Cited by (8)

    View all citing articles on Scopus
    View full text