Table 2.
Reference | Simple | Dataset | Colored Images |
Enhancement | Segmentation | Contribution | Contribution Achieved |
Methods and Tools | No. of Classes |
Accuracy (%) |
Sensitivity (%) |
Specificity (%) |
Precision (%) |
Dice (%) |
||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Number | Type | Free | ||||||||||||||
[100] | Y | 1 | D | Y | N | Y | Y | Mobile Assessment | N | ABCD, MLP | 2 | 88 | 66 | 93 | N/A | N/A |
[101] | N | 1 | A | N | N | Y | Y | Classify lesion Atlas | Y | Gabor filter, ABCD | 2 | 94 | 91.25 | 95.83 | N/A | N/A |
[102] | N | 1 | A | N | Y | Y | Y | Classify lesion Atlas based on the thickness |
Y | logistic regression with ANN | 2 | 64.4 | 55.2 | N/A | N/A | N/A |
[103] | N | 1 | A | N | Y | Y | Y | Increase classification using a novel segmentation method | Y | Median Filter | 2 | N/A | N/A | N/A | N/A | N/A |
1 | D | Y | ||||||||||||||
[104] | N | 3 | A | N | Y | Y | Y | Segment and classify pigmented lesions |
Y | Anisotropic Diffusion, Chan–Vese’s, ABCD, SVM |
2 | 79.01 | N/A | N/A | 80 | N/A |
1 | D | Y | ||||||||||||||
[105] | N | 1 | RS | N | - | - | - | classification | Y | Paraconsistent analysis network | 3 | 93.79 | N/A | N/A | N/A | N/A |
[106] | Y | 1 | D | Y | N | Y | Y | Segmentation | Y | Delaunay Triangulation | 2 | 75.91 | 67.46 | 95.93 | N/A | N/A |
[107] | N | 1 | H | N | Y | N | Y | Segmentation and Classification | Y | Granular Layer, Intensity Profiles | 2 | 92.1 | N/A | 97.6 | 88.1 | N/A |
[108] | Y | 1 | F | N | - | - | - | Comparison of four Classification methods |
Y | KNN with Sequential Scanning selection, KNN with GA, ANN with GA, Adaptive Neuro-Fuzzy Inference |
2 | 94 | N/A | N/A | N/A | N/A |
[109] | Y | 1 | H | N | Y | N | Y | Classification lesions with and without segmentation |
Y | Z-transforms | 2 | 85.18 | 91.66 | 80 | 78.57 | N/A |
[110] | Y | 1 | JPGE | N | Y | N | N | Classification of Psoriasis | Y | PCA, SVM | 2 | 100 | 100 | 100 | N/A | N/A |
[111] | Y | 3 | A | N | N | N | Y | Segmentation | Y | k-means | 2 | N/A | N/A | N/A | N/A | N/A |
[112] | Y | 1 | A | N | Y | N | Y | Classification of skin cancer based on the deep of lesion using 3D reconstruction |
Y | adaptive snake, stereo vision, structure from motion, depth from focus |
8 | 86 | 98 | 99 | N/A | N/A |
1 | D | Y | 3 | |||||||||||||
[113] | N | 2 | D | N | N | N | Y | Classification of melanoma | Y | a self-generated neural network, fuzzy neural networks with backpropagation (BP) neural networks |
2 | 94.17 | 95 | 93.75 | N/A | N/A |
[114] | N | 1 | JPGE | N | Y | Y | Y | Segmentation and Classification | Y | fixed grid wavelet network, D-optimality orthogonal matching pursuit |
2 | 91.82 | 92.61 | 91 | N/A | N/A |
[115] | N | 1 | SI | N | N | N | Y | Classification of the regular and irregular boundary of skin lesion |
Y | 1D time series for Lyapunov exponent and Kolmogorov–Sinai entropy | 2 | 95 | 100 | 92.5 | 86.5 | N/A |
1 | D | Y | ||||||||||||||
[116] | Y | 2 | D | Y | Y | N | Y | Segmentation and classification of lesions to melanoma and Benign | Y | An ant colony, KNN, ANN | 2 | N/A | N/A | N/A | N/A | N/A |
[117] | N | 1 | A | N | Y | N | Y | Segmentation and combine features, classification |
Y | ABCD, SVM | 2 | 90 | 72.5 | 94.4 | N/A | N/A |
2 | D | Y | ||||||||||||||
[118] | N | 2 | D | N | N | Y | Y | Codebook generated from a bag of features |
Y | Histogram of Gradients, Histogram of Lines, Zernike moments |
2 | 92.96 | 96.04 | 84.78 | N/A | N/A |
[119] | N | 1 | D | Y | Y | N | Y | Classification of skin lesion using feature subset selection | Y | Optimum-path forest integrated with a majority voting |
2 | 94.3 | 91.8 | 96.7 | N/A | N/A |
[120] | N | 1 | SIA scope | N | N | N | Y | Analysis of multispectral skin lesion | Y | Box-counting dimension and lacunarity, Hunter score pattern detection, RBF kernel, SVM | 2 | 97 | 59.6 | 97.8 | N/A | N/A |
[121] | N | 2 | D | Y | Y | Y | Y | Automatic segmentation of skin lesion | Y | The grab-Cut algorithm, k-means | 2 | 96.04 | 89 | 98.79 | N/A | 91.39 |
[122] | Y | 1 | CCD | N | Y | N | Y | Classification of skin lesion using a smartphone |
Y | Otsu’s, Minimum Spanning Tree, Color Triangle | 2 | 87.98 | 89.09 | 86.87 | N/A | N/A |
[123] | Y | 1 | D | Y | Y | Y | N | Enhancement and fusion of skin lesion features for classification | Y | Wavelet transform, Curvelet transform, local binary pattern, SVM | 2 | 86.17 | 78.93 | 93.25 | N/A | N/A |
[124] | N | - | - | - | N | Y | Y | Extract fabric characteristics for lesion classification | Y | Gabor filters based on shark smell optimizing method, K-means | - | N/A | N/A | N/A | N/A | N/A |
[125] | N | 3 | D | Y | N | Y | Y | Skin lesion segmentation and recognition based on fused features | Y | ABCD, fuzzy C-means, pair threshold binary decomposition, HOG, linear discriminant analysis, linear regression, complex tree, W-KNN, Naive Bayes, ensemble boosted tree, ensemble subspace discriminant analysis | 2, 3 | 99 | 98.5 | 99 | 98.5 | N/A |
[126] | N | 1 | D | Y | Y | Y | Y | Skin lesion segmentation with several classifiers | Y | 2-D Gabor wavelet, OTSU’s, median filtering, morphological operations, m- mediod classifier, SVM, Gaussian Mixture Modeling | 2 | 97.26 | 96.48 | 96.9 | N/A | N/A |
[127] | N | 3 | D | Y | Y | Y | Y | Classify the selected features of parallel fusion of features after segmentation | Y | Global maxima and minima, uniform and normal distribution, mean, mean deviation, HOG, Harlick method, M-SVM | 2 | 93.2 | 93 | 97 | 93.5 | N/A |
[128] | N | 3 | D | Y/N | N | Y | Y | Optimize features of skin lesion and classify these features | Y | Particle Swarm Optimization, KNN, SVM, decision tree | 2 | N/A | N/A | N/A | N/A | N/A |
[129] | N | 1 | D | Y | Y | Y | Y | Using high discriminative features for melanoma classification | Y | Histogram correction, OTSU’s, corner detection, Gray-level co-occurrence matrix features, Daugman’s rubber sheet model, RUSBoost, linear SVM | 2 | 95 | 95 | 95 | N/A | 92 |
[130] | N | 3 | D | Y | N | Y | Y | Classify melanoma based on the main frequencies from dermoscopic images | Y | ABCD, HU Moments, GLCM, Structural Co-occurrence matrix, MLP, LSSVM, Minimal Learning Machine | 2 | 89.93 | 92.15 | 89.9 | N/A | 91.05 |
[131] | N | 1 | I | N | - | Y | Y | Differentiate infrared spectroscopy of skin lesion | Y | Fourier transform infrared, Morphological information, PCA | 2 | 85 | N/A | N/A | N/A | N/A |
[132] | N | 1 | D | Y | Y | Y | Y | Classifying four skin lesions based on the Gray-level difference method | N | Gray-level difference method, ABCD, SVM | 4 | 100 | 100 | 100 | N/A | N/A |
[133] | N | 1 | CCD | N | Y | Y | Segment skin lesion and using a hybrid classifier to classify these lesions | Y | ABCD, pigment distribution and texture, GLCM, Log-linearized Gaussian mixture neural network, KNN, linear discriminant analysis, LDA, SVM, majority vote | 3 | 98.5 | 95 | 99.5 | N/A | N/A | |
[134] | N | 1 | D | Y | Y | Y | Y | Classify typical and atypical pigment network to diagnose melanoma | Y | Laplacian filter, median filter, polynomial curve fitting, connected component analysis, 2D Gabor filters, Gray-Level co-occurrence Matrix, Pearson product-moment co-relation, Probabilistic SVM, ANN | 2 | 86.7 | 84.6 | 88.7 | N/A | N/A |
[135] | N | 1 | LIBS spectra | N | - | Y | Y | Classification of skin tissue | Y | laser-induced breakdown spectroscopy, PCA, KNN, SVM | 2 | 76.84 | 74.2 | 86.9 | N/A | N/A |
[136] | Y | 2 | A | N | N | Y | Y | Classification of melanoma versus nevi by correlation bias reduction | Y | 2D wavelet packet decomposition, SVM recursive feature elimination (SVM -RFE) | 2 | 98.28 | 97.63 | 100 | N/A | N/A |
2 | D | Y | ||||||||||||||
[137] | Y | 1 | D | N | Y | Y | Y | Classification of lesions into melanoma and nevi | Y | Gaussian Filter, KNN, SVM | 2 | 96 | 97 | 96 | 97 | N/A |
[138] | Y | 1 | A | N | Y | N | Y | Using of blue-whitish structure for melanoma classification | Y | multiple instance learning (MIL) paradigm, Markov network, SVM | 2 | 84.5 | 74.42 | 87.9 | 61.54 | N/A |
1 | D | Y | ||||||||||||||
[139] | N | 1 | A | N | Y | N | Y | Classify lesion into melanoma or nevi based of color features | Y | K-means, pixel-based classification | 2 | 89.42 | N/A | N/A | N/A | N/A |
[140] | Y | 1 | D | Y | Y | Y | Segmentation and classification of skin lesions | Y | Hough Transform, ABCD, SP-SIFT, LF-SLIC region labeling | 2 | 96 | N/A | N/A | N/A | 93.8 | |
[141] | Y | 2 | D | Y | Y | Y | Y | Detect the boundaries of lesions to classify into melanoma and nevi | Y | Kullback-Leibler divergence, local binary patterns, SVM, KNN | 2 | 80.7 | N/A | N/A | N/A | N/A |
[142] | N | 1 | JPGE | N | Y | Y | Y | A QuadTree-based melanoma detection system based on color | Y | hybrid thresholding method, adaptive histogram thresholding, Euclidean distance transform, QuadTree, Kolmogorov-Smirnov, SVM, ANN, LDA, random forests | 2 | 73.8 | 75.7 | 73.3 | N/A | N/A |
[143] | Y | 1 | D | Y | N | Y | Y | Segmentation and classification of skin lesion | Y | Median Filter, watershed segmentation, ABCD, GLCM, KNN, RF, SVM | 2 | 89.43 | 91.15 | 87.71 | N/A | N/A |
[144] | Y | 1 | D | Y | N | Y | Y | Segmentations of Enhanced dermoscopic lesion images | Y | wavelet transform, morphological operations, Gray Thresholding, Cohen–Daubechies–Feauveau biorthogonal wavelet, Active contour, Color enhancement, Adaptive thresholding, Gradient vector flow | 2 | 93.87 | N/A | N/A | N/A | 92.72 |
[145] | Y | 1 | D | Y | N | Y | Y | Three distinct features to classify melanoma | N | ABCD, Otsu, Chan–Vese, Dull-Razor, ANN | 2 | 98.2 | 98 | 98.2 | N/A | N/A |
[146] | N | 2 | A | N | N | Y | Y | Classification three lesions based on shape, fractal dimension, texture, and color features | Y | recursive feature elimination, GLCM, fractal-based regional texture analysis, SVM, RBF | 3 | 98.99 | 98.28 | 98.48 | N/A | 91.42 |
2 | D | Y | ||||||||||||||
[147] | N | 2 | A | N | Y | N | N | Extraction of features using frequency domain analysis and classify these features | Y | Cross spectrum-based feature extraction, Spatial feature extraction, SVM-RFE with CBR | 4 | 98.72 | 98.89 | 98.83 | N/A | N/A |
2 | D | Y | ||||||||||||||
[148] | Y | 1 | D | N | Y | N | N | Improve bag of dense features to Classify skin lesions | Y | Gradient Location, Orientation Histogram, color features, SVM | 2 | 78 | N/A | N/A | N/A | N/A |
[149] | N | 2 | D | Y | N | Y | Y | CAD system for clinical assist | N | chroma based deformable models, speed function, Chan-Vese, Wilcoxon Rank Sum statistics, Discrete Wavelet Transform, Asymmetry, and Compactness Index, SVM | 2 | 88 | 95 | 82 | N/A | N/A |
[150] | Y | 2 | D | Y | N | Y | Y | Segmentation of skin lesions using fuzzy pixels classification and histogram thresholding | Y | fuzzy classification, histogram thresholding | 2 | 88.4 | 86.9 | 92.3 | N/A | 76 |
[151] | N | 1 | D | Y | Y | Y | Y | Lesion classification based on feature similarity measurement for codebook learning in the bag-of-features model | Y | Codebook learning, k-means, color histogram, scale-invariant feature transform (SIFT) | 2 | 82 | 80 | 83 | N/A | N/A |
1 | C | Y | ||||||||||||||
[152] | Y | 1 | D | Y | Y | Y | Y | Segmentation and classification of skin lesions using Kernel sparse representation | Y | Sparse coding, kernel dictionary, K-SVD | 3 | 91.34 | 93.17 | 91.48 | N/A | 91.25 |
[153] | N | 1 | D | N | N | Y | Y | Improve skin lesion classification using borderline characteristics | Y | Gradient-based Histogram Thresholding, Local Binary Patterns Clustering, Euclidean distance, Discrete Fourier Transform spectrum (DCT), power spectral density (PSD), SVM, Feedforward Neural Network (FNN) | 2 | 91 | 68 | 96 | N/A | N/A |
1 | C | Y | ||||||||||||||
[154] | Y | 1 | D | N | Y | N | Y | Multi-resolution-Tract CNN | N | AlexNet, GPU | 10 | 79.5 | N/A | N/A | N/A | N/A |
[155] | Y | 1 | D | Y | Y | Y | Y | Automatic segmentation and classification for skin lesions | Y | CNN with 50 layers, residual learning, SoftMax, SVM, Augmentation, GPU | 2 | 94 | N/A | N/A | N/A | N/A |
[156] | N | 1 | D | Y | N | Y | Y | Classification of segmented skin lesions | Y | U-Net, Sparse coding, Deep Residual Network (DRN), Augmentation | 2 | 76 | 82 | 62 | N/A | N/A |
[157] | Y | 1 | D | Y | N | Y | Y | Segmentation of skin lesions using deep learning | Y | fully convolutional networks (FCN), VGG, Augmentation | 2 | N/A | N/A | N/A | N/A | 89.2 |
[158] | Y | 1 | D | Y | Y | Y | Y | Automatic Skin Lesion Segmentation Using CNN | Y | FCN with 19 layers, Jaccard Distance, Augmentation, GPU | 2 | 95.5 | 91.8 | 96.6 | N/A | 92.2 |
[159] | Y | 3 | D | Y | Y | Y | N | Detection of melanoma using CNN and regularized fisher framework | Y | ResNet50, transfer learning, SoftMax, SVM, Augmentation, GPU | 2 | 78.3 | 35 | 88.8 | N/A | N/A |
1 | C | Y | ||||||||||||||
[160] | N | 1 | D | Y | N | Y | Y | Evaluation of skin lesion using Levenberg neural networks and stacked autoencoders clustering | Y | ABCD, morphological analysis, Levenberg–Marquardt neural network | 2 | N/A | 98 | 98 | N/A | N/A |
[161] | N | 1 | D | Y | Y | N | N | Classify limited and imbalanced skin lesion images | N | Adversarial Autoencoder with 19 layers, Augmentation, GPU | 2 | N/A | N/A | 83 | N/A | N/A |
[162] | Y | 1 | D | Y | Y | N | N | Ensemble different CNN for skin lesion classification | Y | GoogLeNet, AlexNet, ResNet, VGGNet, Sum of the probabilities, Product of the possibilities, Simple majority voting (SMV), Sum of the maximal probabilities (SMP), Weighted ensemble of CNN, Augmentation, GPU | 3 | 86.6 | 55.6 | 78.5 | N/A | N/A |
[163] | N | 1 | D | Y | Y | N | N | Skin lesion analysis using Multichannel ResNet | Y | Ensemble multi-ResNet50, ANN, concatenated Fully Connected Layer, Augmentation, GPU | 3 | 82.4 | N/A | N/A | N/A | N/A |
[164] | N | 1 | A | N | Y | Y | Y | Classify skin lesions based on border thickness | Y | GoogleNet, Breslow index | 5 | 66.2 | 89.19 | 85 | N/A | N/A |
[165] | N | 1 | D | Y | Y | Y | Y | Classify lesion using fused features that extracted from deep learning and image processing | Y | ResNet50, Telangiectasia Vessel Detection Algorithm, transfer learning, GPU | 5 | N/A | N/A | N/A | N/A | N/A |
[166] | Y | 1 | D | Y | Y | N | N | Classify skin lesions over IoT using deep learning | Y | CNN with nine layers, IoT, | 7 | 81.4 | N/A | N/A | N/A | N/A |
[168] | Y | 3 | D | Y | Y | Y | N | Skin lesion classification using depthwise separable residual CNN | Y | Non-local means filter, contrast-limited adaptive histogram equalization, discrete Wavelet transforms, depthwise separable residual DCNN | 2 | 99.5 | 99.31 | 100 | N/A | N/A |
1 | C | Y | ||||||||||||||
[169] | Y | 1 | D | Y | Y | Y | Y | Classify skin lesion using Attention Residual Learning | Y | Attention residual learning CNN, ResNet50, transfer learning, Augmentation, GPU | 3 | N/A | N/A | N/A | N/A | N/A |
[170] | Y | 1 | D | Y | Y | N | N | Classify skin lesion using CNN with novel regularization method | Y | CNN 7 layers, standard deviation of the weight matrix, GPU | 2 | 97.49 | 94.3 | 93.6 | N/A | N/A |
[171] | N | 1 | D | Y | N | Y | Y | Classify skin lesion by Incorporating the knowledge of dermatologists to CNN | Y | ResNet50, DermaKNet, Modulation Block, asymmetry block, AVG layer, Polar AVG layer. | 3 | 91.7 | N/A | 65.2 | N/A | N/A |
[172] | Y | 1 | - | - | Y | Y | Y | Classify skin lesions based on the novel 7-point melanoma checklist using Multitask CNN | Y | Multitask CNN, 7-point melanoma checklist, Augmentation, GPU | 3 | 87.1 | 77.3 | 89.4 | 63 | N/A |
[173] | Y | 1 | D | Y | Y | N | N | Classify skin lesion using Aggregated CNN | Y | ResNet50, ResNet101, fisher vector (FV), SVM, Chi-squared kernel, transfer learning, Augmentation, GPU | 2 | 86.81 | N/A | N/A | N/A | N/A |
[174] | Y | - | - | - | Y | N | N | Using CNN as a feature extractor for skin lesion images and classify these features | Y | Alex-Net, ECOC SVM, transfer learning | 4 | 94.2 | 97.83 | 90.74 | N/A | N/A |
[175] | Y | 1 | D | Y | Y | N | N | Classification of skin neoplasms using CNN and transfer learning with web and mobile application | N | Inception V3 (GoogleNet), transfer learning, Augmentation, GPU | - | 91 | N/A | N/A | N/A | N/A |
[176] | N | 1 | C | N | Y | N | Y | An application used through the cloud to classify diseases of face skin | Y | LeNet-5, AlexNet and VGG16, transfer learning, Augmentation, GPU | 5 | N/A | N/A | N/A | N/A | N/A |
[177] | Y | 2 | D | Y | Y | Y | Y | Skin lesion classification using 4 CNNs and ensembling of the final classification results | Y | AlexNet, VGG, ResNet-18, ResNet-101, SVM, MLP, random forest, transfer learning, Augmentation, GPU | 3 | 87.7 | 85 | 73.29 | N/A | N/A |
[178] | Y | 1 | D | Y | Y | N | N | Compare the ability of deep learning model to classify skin lesions with expert dermatologists | Y | ResNet50, local outlier factor, transfer learning, GPU | 2 | N/A | 87.5 | 60 | N/A | N/A |
[179] | N | 1 | D | N | Y | Y | Y | Evolving the deep learning model | PSO, hybrid learning PSO, Firefly Algorithm, spiral research action, probability distributions, crossover, mutation, K-Means, VGG16, Augmentation, GPU | 2 | 73.76 | N/A | N/A | N/A | N/A | |
2 | D | Y | ||||||||||||||
[181] | Y | 3 | D | Y | Y | Y | Y | Combine and expand current segmentation CNN to enhance the classification of skin lesions | Y | U-Net, ResNet34, LinkNet34, LinkNet152, fine-tuning, PyTorch, transfer learning, Augmentation, GPU, Jaccard-loss, | 2 | N/A | N/A | N/A | N/A | N/A |
[182] | N | 1 | D | Y | N | Y | Y | skin lesion segmentation based using geodesic morphological active contour | Y | Gaussian filter, Otsu’s threshold, deformable models, partial differential equation, Mathematical morphology, active geodesic contour, neural network, deep learning, statistical region merging (SRM), | 2 | 94.59 | 91.72 | 97.99 | N/A | 89 |
[183] | Y | 1 | - | - | Y | N | N | Erythema migrans and the other confounding lesions of skin using | Y | ResNet50, Keras, TensorFlow, fine-tuning, transfer learning, Augmentation, GPU | 4 | 86.53 | 76.4 | 75.96 | N/A | 92.09 |
[184] | Y | 1 | D | Y | Y | N | N | Comparing between the CNN and 112 dermatologists for skin lesion detection | Y | ResNet50, fine-tuning, transfer learning, Augmentation, GPU | 5 | N/A | 56.5 | 98.2 | N/A | N/A |
[185] | Y | 2 | D | Y | Y | Y | Y | Segmentation of skin lesions by ensemble the segmentation output of 2 CNN | Y | DEEPLABV3+, Mask R-CNN, ABCD, fine-tuning, transfer learning, Augmentation, GPU | 2 | 94.08 | 89.93 | 95 | N/A | N/A |
[186] | Y | 1 | C | Y | N | Y | Y | Proposing an algorithm that able to train CNN with limited data | Y | Inception V3 (GoogleNet), PECK, SCIDOG, SVM, RF, fine-tuning, transfer learning, Augmentation, GPU | 2 | 91 | 92 | 93 | N/A | 90.7 |
[187] | Y | 1 | C | N | Y | N | N | Classify skin disease of faces using Euclidean space to compute L-2 distance between images | Y | ResNet152, InceptionResNet-V2, fine-tuning, Euclidean space, L-2 distance, transfer learning, Augmentation, GPU | 4 | 87.42 | 97.04 | 97.23 | N/A | N/A |
[188] | Y | 1 | D | Y | Y | Y | Y | Neural Architecture Search to increase the size of the network based on the dataset size | Y | VGG8, VGG11, VGG16, 5-fold validation, Neural Architecture Search (NAS), hill-climbing, transfer learning, Augmentation, GPU | 2 | 77 | N/A | N/A | N/A | N/A |
[189] | Y | 2 | D | Y | Y | Y | Y | Skin lesion segmentation and pixel-wise classification using encoder and decoder network | Y | CNN, encoder-decoder deep network with skip connection, softmax, transfer learning, Augmentation, GPU | 2 | 95 | 97 | 96 | N/A | N/A |
[190] | Y | 2 | D | Y | Y | Y | Y | Multitasks DCNN for skin lesion segmentation, detection, and classification | Y | Multitask DCNN, Jaccard distance, focal loss, Augmentation, GPU | 2 | 95.9 | 83.1 | 98.6 | N/A | 95 |
[191] | Y | 1 | D | Y | Y | Y | Y | Light Lightweight CNN for skin lesion segmentation and classification | Y | Lightweight CNN, MobileNet, DenseNet, U-Net, focal loss, fine-tune, transfer learning, Augmentation, GPU | 2 | 96.2 | 93.4 | 97.4 | N/A | 88.9 |
[192] | N | 1 | D | Y | Y | Y | N | Enhancement and classification of dermoscopic skin images | Y | StyleGANs, 43 CNNs (ResNet50, VGG11, VGG13, AlexNet, SENet, etc.) max voting, fine-tune, transfer learning, Augmentation, GPU | 8 | 99.5 | 98.3 | 99.6 | N/A | 92.3 |
[193] | Y | 1 | D | Y | Y | Y | Y | Skin lesion classification based on CNN | Y | Deep-class CNN, Augmentation, GPU | 2 | 75 | 73 | 78 | N/A | N/A |
[194] | Y | 2 | D | Y | Y | Y | Y | Skin lesion segmentation using encoder-decoder FCN | Y | FCN, GPU | 3 | 96.92 | 96.88 | 95.31 | N/A | N/A |
[195] | Y | 3 | D | Y | Y | N | Y | Classify skin melanoma by extracting ROI and CNNS | Y | AlexNet, ResNet101, GoogleNet, Multiclass SVM, SoftMax, Histogram based windowing process, hierarchical clustering, fine-tune, transfer learning, Augmentation, GPU | 3 | 98.14 | 97.27 | 98.60 | N/A | 88.64 |
1 | C | Y | ||||||||||||||
[196] | Y | 3 | D | Y | N | Y | Y | The fusion of extracted deep features of a skin lesion for classification | Y | Biorthogonal 2-D wavelet transform, Otsu algorithm, Alex and VGG-16, PCA, fusion, fine-tune, transfer learning, Augmentation, GPU | 2 | 99.9 | 99.5 | 99.6 | N/A | N/A |
[197] | N | 3 | D | Y | Y | Y | Y | Ensemble multiscale and multi-CNN network | Y | EfficientNetB0, EfficientNetB1, SeResNeXt-50, fusion, fine-tune, transfer learning, Augmentation, GPU | 7 | 96.3 | N/A | N/A | 91.3 | 82 |
[198] | Y | 2 | D | Y | Y | Y | Y | Classification of skin lesions by multiclass multilevel using traditional machine learning and transfer learning | Y | K-means, Otsu’s thresholding, GLCM, ANN. k-fold validation, AlexNet, fine-tune, transfer learning, Augmentation, GPU | 4 | 93.02 | 87.87 | 98.17 | 97.96 | N/A |
[199] | N | 2 | D | Y | N | Y | Y | Optimized algorithm for weight selection to minimize the output of the network | Y | The bubble-net mechanism, Whale optimization algorithm (WOA), Lévy Flight Mechanism, genetic algorithm, shark smell optimization (SSO), world cup optimization algorithm, grasshopper optimization algorithm (GOA), particle swarm optimization algorithm (PSO), LeNet, fine-tune, transfer learning, GPU | 2 | N/A | N/A | N/A | N/A | N/A |
[200] | Y | 3 | D | Y | N | Y | Y | semantic skin lesion segmentation with parameters reducing | Y | U-Net, FCN8s, DSNet, Augmentation, GPU | 7 | N/A | 87.5 | 95.5 | N/A | N/A |
[201] | N | 3 | D | Y | Y | Y | Y | Different CNN network integration for segmentation and multiple classification stage | Y | Inception-v3, ResNet-50, Inception-ResNet-v2, and DenseNet-201 | 7 | 89.28 | 81 | 87.16 | N/A | 81.82 |
[202] | Y | 3 | D | Y | Y | Y | Y | Skin lesion segmentation based on CNN | Y | CIElab, FCN, U-Net, Augmentation, GPU | 2 | N/A | N/A | N/A | N/A | 87.1 |
[205] | Y | 1 | D | Y | Y | N | N | Classify the challenging dataset ISIC2018 | Y | AlexNet, 10-fold cross-validation, fine-tune, transfer learning, Augmentation, GPU | 7 | 92.99 | 70.44 | 96 | 62.78 | N/A |
[204] | Y | 1 | D | Y | Y | N | N | Classify the challenging dataset ISIC2019 | Y | GoogleNet, Similarity score, bootstrap weighted SVM classifier, SoftMax, fine-tune, transfer learning, Augmentation, GPU | 9 | 98.70 | 95.6 | 99.27 | 95.06 | N/A |