A basis function feature‐based approach for skin lesion discrimination in dermatology dermoscopy images

Background: Skin lesion color is an important feature for diagnosing malignant melanoma. New basis function correlation features are proposed for discriminating malignant melanoma lesions from benign lesions in dermoscopy images. The proposed features are computed based on correlating the luminance histogram of melanoma or benign labeled relative colors from a specified portion of the skin lesion with a set of basis functions. These features extend previously developed statistical and fuzzy logic‐based relative color histogram analysis techniques for automated mapping of colors representative of melanoma and benign skin lesions from a training set of lesion images. Methods: Using the statistical and fuzzy logic‐based approaches for relative color mapping, melanoma and benign color features are computed over skin lesion region of interest, respectively. Luminance histograms are obtained from the melanoma and benign mapped colors within the lesion region of interest and are correlated with a set of basis functions to quantify the distribution of colors. The histogram analysis techniques and feature calculations are evaluated using a data set of 279 malignant melanomas and 442 benign dysplastic nevi images. Results: Experimental test results showed that combining existing melanoma and benign color features with the proposed basis function features found from the melanoma mapped colors yielded average correct melanoma and benign lesion discrimination rates as high as 86.45% and 83.35%, respectively. Conclusions: The basis function features provide an alternative approach to melanoma discrimination that quantifies the variation and distribution of colors characteristic of melanoma and benign skin lesions.     

A relative color approach to color discrimination for malignant melanoma detection in dermoscopy images

BACKGROUND: Skin lesion color is an important feature for diagnosing malignant melanoma. In previous research, skin lesion color was investigated for discriminating malignant melanoma lesions from benign lesions in clinical images. Colors characteristics of melanoma were determined using color histogram analysis over a training set of images. Percent melanoma color and color clustering ratio features were used to quantify the presence of melanoma-colored pixels within skin lesions for skin lesion discrimination. METHODS: In this research, the relative color histogram analysis technique is used to evaluate skin lesion discrimination based on color feature calculations in different regions of the skin lesion in dermoscopy images. The histogram analysis technique is examined for varying training set sizes from the set of 113 malignant melanomas and 113 benign dysplastic nevi images. RESULTS: Experimental results show improved discrimination capability for feature calculations focused in the interior lesion region. Recognition rates for malignant melanoma and dysplastic nevi as high as 87.7% and 74.9%, respectively, are observed for the color clustering ratio computed using the outer 75% uniformly distributed area with a 10% offset within the boundary. CONCLUSIONS: Experimental results appear to indicate that the melanoma color feature information is located in the interior of the lesion, excluding the 10% central-most region. The techniques presented here including the use of relative color and the determination of benign and malignant regions of the relative color histogram may be applicable to any set of images of benign and malignant lesions.     

Colour analysis of skin lesion regions for melanoma discrimination in clinical images

Background: Skin lesion colour is an important feature for diagnosing malignant melanoma. Colour histogram analysis over a training set of images has been used to identify colours characteristic of melanoma, i.e., melanoma colours. A percent melanoma colour feature defined as the percentage of the lesion pixels that are melanoma colours has been used as a feature to discriminate melanomas from benign lesions.
Methods: In this research, the colour histogram analysis technique is extended to evaluate skin lesion discrimination based on colour feature calculations in different regions of the skin lesion. The colour features examined include percent melanoma colour and a novel colour clustering ratio. Experiments are performed using clinical images of 129 malignant melanomas and 129 benign lesions consisting of 40 seborrheic keratoses and 89 nevocellular nevi.
Results: Experimental results show improved discrimination capability for feature calculations focused in the lesion boundary region. Specifically, correct melanoma and benign recognition rates are observed as high as 89 and 83%, respectively, for the percent melanoma colour feature computed using only the outermost, uniformly distributed 10% of the lesion's area.
Conclusions: The experimental results show for the features investigated that the region closest to the skin lesion boundary contains the greatest colour discrimination information for lesion screening. Furthermore, the percent melanoma colour feature consistently outperformed the colour clustering ratio for the different skin lesion regions examined. The clinical application of this result is that clustered colours appear to be no more significant than colours of arbitrary distribution within a lesion.     

Differentiation of melanoma from benign mimics using the relative‐color method

Background: Previous studies have successfully classified 86% of malignant melanomas using a relative‐color segmentation method, by feature extraction from photographic images in the automatic identification of skin tumors. These studies were extended by applying the relative‐color method to dermoscopic images of melanoma grouped with melanoma in situ and clark nevus lesions in dermoscopic images allow more control over lighting variations, which contribute to lesion misclassification. Dermoscopic images then enable a more detailed examination of the structure of skin lesions, provide much more structural detail within lesions, and contain visual information that cannot be seen in photographic images. This present work extends the previous studies by applying relative‐color feature extraction to dermoscopic images to differentiate among melanoma, seborrheic keratoses and Reed/Spitz nevi. Objective: To develop a method for automatically differentiating among malignant melanoma, seborrheic keratoses and Reed/Spitz nevi, using digitized, color, dermoscopic images. Methods: Images underwent preprocessing, tumor segmentation, feature extraction and tumor classification. The relative‐color method was used in the segmentation stage. Classification was accomplished by taking the inner products of model tumor feature vectors with test‐image tumor vectors followed by the nearest‐neighbor classification method. Results: The classification rates of melanoma, seborrheic keratoses and Reed/Spitz nevi images mixed together, were 60%, 58.3% and 80%, respectively. Classification of melanoma and Reed/Spitz nevi mixed, were 70% and 90%, respectively. Classification rates were the best when melanoma was being differentiated from seborrheic keratoses. These rates were 100% and 88.9%, respectively. Conclusion: Dermoscopic rather than photographic images were preprocessed, using a hair‐removal technique. They were then converted to relative‐color images, which were segmented using the principal components transform and median split, followed by morphological filtering. After processing, the multi‐dimensional tumor feature space described herein was used to differentiate the tumors. The high success rates for differentiating seborrheic keratoses from melanoma show that the use of dermoscopic images has a strong promise in enabling prescreening, as well as automated assistance and significant improvement in tumor diagnosis in clinics.     

Extraction of skin lesion texture features based on independent component analysis

Background/purpose: During the recent years, many diagnostic methods have been proposed aiming at early detection of malignant melanoma. The texture of skin lesions is an important feature to differentiate melanoma from other types of lesions, and different techniques have been designed to quantify this feature. In this paper, we discuss a new approach based on independent component analysis (ICA) for extraction of texture features of skin lesions in clinical images.
Methods: After preprocessing and segmentation of the images, features that describe the texture of lesions and show high discriminative characteristics are extracted using ICA, and then these features, along with the color features of the lesions, are used to construct a classification module based on support vector machines for the recognition of malignant melanoma vs. benign nevus.
Results: Experimental results showed that combining melanoma and nevus color features with proposed ICA-based texture features led to a classification accuracy of 88.7%.
Conclusion: ICA can be used as an effective tool for quantifying the texture of lesions.     

Colour histogram analysis for melanoma discrimination in clinical images

Background: Malignant melanoma, the most deadly form of skin cancer, has a good prognosis if treated in the curable early stages. Colour provides critical discriminating information for the diagnosis of malignant melanoma.
Methods: This research introduces a three-dimensional relative colour histogram analysis technique to identify colours characteristic of melanomas and then applies these 'melanoma colours' to differentiate benign skin lesions from melanomas. The relative colour of a skin lesion is determined based on subtracting a representative colour of the surrounding skin from each lesion pixel. A colour mapping for 'melanoma colours' is determined using a training set of images. A percent melanoma colour feature, defined as the percentage of the lesion pixels that are melanoma colours, is used for discriminating melanomas from benign lesions. The technique is evaluated using a clinical image data set of 129 malignant melanomas and 129 benign lesions consisting of 40 seborrheic keratoses and 89 nevocellular nevi.
Results: Using the percent melanoma colour feature for discrimination, experimental results yield correct melanoma and benign lesion discrimination rates of 84.3 and 83.0%, respectively.
Conclusions: The results presented in this work suggest that lesion colour in clinical images is strongly related to the presence of melanoma in that lesion. However, colour information should be combined with other information in order to further reduce the false negative and false positive rates.     

Detection of asymmetric blotches (asymmetric structureless areas) in dermoscopy images of malignant melanoma using relative color

BACKGROUND: Dermoscopy, also known as dermatoscopy or epiluminescence microscopy (ELM), is a non-invasive, in vivo technique, which permits visualization of features of pigmented melanocytic neoplasms that are not discernable by examination with the naked eye. One prominent feature useful for melanoma detection in dermoscopy images is the asymmetric blotch (asymmetric structureless area). METHOD: Using both relative and absolute colors, blotches are detected in this research automatically by using thresholds in the red and green color planes. Several blotch indices are computed, including the scaled distance between the largest blotch centroid and the lesion centroid, ratio of total blotch areas to lesion area, ratio of largest blotch area to lesion area, total number of blotches, size of largest blotch, and irregularity of largest blotch. RESULTS: The effectiveness of the absolute and relative color blotch features was examined for melanoma/benign lesion discrimination over a dermoscopy image set containing 165 melanomas (151 invasive melanomas and 14 melanomas in situ) and 347 benign lesions (124 nevocellular nevi without dysplasia and 223 dysplastic nevi) using a leave-one-out neural network approach. Receiver operating characteristic curve results are shown, highlighting the sensitivity and specificity of melanoma detection. Statistical analysis of the blotch features are also presented. CONCLUSION: Neural network and statistical analysis showed that the blotch detection method was somewhat more effective using relative color than using absolute color. The relative-color blotch detection method gave a diagnostic accuracy of about 77%.     

Combination of features from skin pattern and ABCD analysis for lesion classification

BACKGROUND/PURPOSE: It is known that the standard features for lesion classification are ABCD features, that is, asymmetry, border irregularity, colour variegation and diameter of lesion. However, the observation that skin patterning tends to be disrupted by malignant but not by benign skin lesions suggests that measurements of skin pattern disruption on simply captured white light optical skin images could be a useful contribution to a diagnostic feature set. Previous work using both skin line direction and intensity for lesion classification was encouraging. But these features have not been combined with the ABCD features. This paper explores the possibility of combing features from skin pattern and ABCD analysis to enhance classification performance. METHODS: The skin line direction and intensity were extracted from a local tensor matrix of skin pattern. Meanwhile, ABCD analysis was conducted to generate six features. They were asymmetry, border irregularity, colour (red, green and blue) variegations and diameter of lesion. The eight features of each case were combined using a principal component analysis (PCA) to produce two dominant features for lesion classification. RESULTS: A larger set of images containing malignant melanoma (MM) and benign naevi were processed as above and the scatter plot in a two-dimensional dominant feature space showed excellent separation of benign and malignant lesions. An ROC (receiver operating characteristic) plot enclosed an area of 0.94. CONCLUSIONS: The classification results showed that the individual features have a limited discrimination capability and the combined features were promising to distinguish MM from benign lesion.     

Automatic lesion boundary detection in dermoscopy images using gradient vector flow snakes

BACKGROUND: Malignant melanoma has a good prognosis if treated early. Dermoscopy images of pigmented lesions are most commonly taken at x 10 magnification under lighting at a low angle of incidence while the skin is immersed in oil under a glass plate. Accurate skin lesion segmentation from the background skin is important because some of the features anticipated to be used for diagnosis deal with shape of the lesion and others deal with the color of the lesion compared with the color of the surrounding skin. METHODS: In this research, gradient vector flow (GVF) snakes are investigated to find the border of skin lesions in dermoscopy images. An automatic initialization method is introduced to make the skin lesion border determination process fully automated. RESULTS: Skin lesion segmentation results are presented for 70 benign and 30 melanoma skin lesion images for the GVF-based method and a color histogram analysis technique. The average errors obtained by the GVF-based method are lower for both the benign and melanoma image sets than for the color histogram analysis technique based on comparison with manually segmented lesions determined by a dermatologist. CONCLUSIONS: The experimental results for the GVF-based method demonstrate promise as an automated technique for skin lesion segmentation in dermoscopy images.     

In vivo reflectance confocal microscopy: automated diagnostic image analysis of melanocytic skin tumours

Background In vivo reflectance confocal microscopy (RCM) has been shown to be a valuable imaging tool in the diagnosis of melanocytic skin tumours. However, diagnostic image analysis performed by automated systems is to date quite rare. Objectives In this study, we investigated the applicability of an automated image analysis system using a machine learning algorithm on diagnostic discrimination of benign and malignant melanocytic skin tumours in RCM. Methods Overall, 16 269 RCM tumour images were evaluated. Image analysis was based on features of the wavelet transform. A learning set of 6147 images was used to establish a classification tree algorithm and an independent test set of 10 122 images was applied to validate the tree model (grouping method 1). Additionally, randomly generated ‘new’ learning and test sets, tumour images only and different skin layers were evaluated (grouping method 2, 3 and 4). Results The classification tree analysis correctly classified 93.60% of the melanoma and 90.40% of the nevi images of the learning set. When the classification tree was applied to the independent test set 46.71 ± 19.97% (range 7.81–83.87%) of the tumour images in benign melanocytic skin lesions were classified as ‘malignant’, in contrast to 55.68 ± 14.58% (range 30.65–83.59%; t‐test: P < 0.036) in malignant melanocytic skin lesions (grouping method 1). Further investigations could not improve the results significantly (grouping method 2, 3 and 4). Conclusions The automated RCM image analysis procedure holds promise for further investigations. However, to date our system cannot be applied to routine skin tumour screening.     

Tissue counter analysis of tissue components in skin biopsies: evaluation using CART (Classification and Regression Trees)

In tissue counter analysis, complex histologic sections are overlaid with regularly distributed measuring masks of equal size and shape, and the digital contents of each mask (or tissue element) are evaluated by gray level, color, and texture parameters. In this study, the feasibility of tissue counter analysis and classification and regression trees for the quantitative evaluation of skin biopsies was assessed. From 100 randomly selected skin biopsies, a learning set of tissue elements was created, differentiating between cellular elements, collagenous elements of the reticular dermis, fatty elements and other tissue components. Classification and regression trees based on the learning set were used to automatically classify tissue elements in samples of normal skin, benign common nevi, malignant melanoma, molluscum contagiosum, seborrheic keratosis, epidermoid cysts, basal cell carcinoma, and scleroderma. The procedure yielded reproducible assessments of the relative amounts of tissue components in various diagnostic groups. Furthermore, a reliable diagnostic separation of molluscum contagiosum versus normal skin and epidermal cysts, benign common nevi versus malignant melanoma, and seborrheic keratosis versus basal cell carcinoma was possible. Tissue counter analysis combined with classification and regression trees may be a suitable approach to the fully automated analysis of histologic sections of skin biopsies.     

Melanoma and seborrheic keratosis differentiation using texture features

Purpose: To explore texture features in two-dimensional images to differentiate seborrheic keratosis from melanoma.
Methods: A systematic approach to consistent classification of skin tumors is described. Texture features, based on the second-order histogram, were used to identify the features or a combination of features that could consistently differentiate a malignant skin tumor (melanoma) from a benign one (seborrheic keratosis). Two hundred and seventy-one skin tumor images were separated into training and test sets for accuracy and consistency. Automatic induction was applied to generate classification rules. Data analysis and modeling tools were used to gain further insight into the feature space.
Result and Conclusions: In all, 85–90% of seborrheic keratosis images were correctly differentiated from the malignant skin tumors. The features correlation_average, correlation_range, texture_energy_average and texture_energy_range were found to be the most important features in differentiating seborrheic keratosis from melanoma. Over-all, the seborrheic keratosis images were better identified by the texture features than the melanoma images.     

Using 3D differential forms to characterize a pigmented lesion in vivo

Background/purpose: After the formulation of ABCD rules, many new feature extraction methods are emerging to describe the asymmetry, border irregularity, color variation and diameter of malignant melanoma. In this paper, a new research direction orthogonal to ABCD rules that characterizes 3D local disruption of skin surfaces to realize automatic recognition of melanoma is described.
Methods: This paper examines 3D differential forms of skin surfaces to characterize the local geometrical properties of melanoma. Firstly, 3D data of skin surfaces are obtained using a photometric stereo device. Then differential forms of lesion surfaces are determined to describe the geometrical texture patterns involved. Using only these geometrical features, a simple least-squared error-based linear classifier can be constructed to realize the classification of malignant melanomas and benign lesions.
Results: As with the 3D data of 35 melanoma and 66 benign lesion samples collected from local pigmented lesion clinics, the optimal sensitivity and specificity of the constructed linear classifier are 71.4% and 86.4%, respectively. The total area enclosed by the corresponding receiver operating characteristics curve is 0.823.
Conclusion: This study indicates that differential forms obtained from 3D data are very promising in characterizing melanoma. Combining these features with other skin features such as border irregularity and color variation might further improve the accuracy and reliability of the automatic diagnosis of melanoma.     

Skin pattern analysis for lesion classification using local isotropy

Background/purpose: The observation that skin pattern tends to be disrupted by malignant skin lesions, but not by benign ones suggests that measurements of skin pattern disruption on simply captured white light optical clinical (WLC) skin images could be a useful contribution to a diagnostic feature set. Previous work, which generated a flow field of skin pattern using a measurement of local line direction and intensity, was encouraging. The aim of this paper is to investigate the possibility of extracting new features using local isotropy metrics to quantify the skin pattern disruption. Methods: The skin pattern was extracted from WLC skin images by high‐pass filtering. A local tensor matrix was computed. The local isotropy was measured by the condition number of the local tensor matrix. The difference of this measure over the lesion and normal skin areas, combined with the local line direction and the ABCD features, was used as a lesion classifier. Results: A set of images of malignant melanoma and benign naevi was analysed. A one‐dimensional scatter plot showed the potential of a local isotropy metric, showing an area of 0.70 under the receiver operating characteristic (ROC) curve. A two‐dimensional scatter plot, combined with the local line direction, indicated enhancement of the classification performance, showing an area of 0.89 under the ROC curve. A three‐dimensional scatter plot combined with the local line direction and the ABCD features, using principal component analysis, demonstrated excellent separation of benign and malignant lesions. An ROC plot for this case enclosed an area of 0.96. Conclusion: The experimental results show that the local isotropy metric has a potential to increase lesion classifier accuracy. Combined with the local line direction and the ABCD features, it is very promising as a method to distinguish malignant melanoma from benign lesions.     

Enhancement of lesion classification using divergence, curl and curvature of skin pattern

BACKGROUND/PURPOSE: The observation that skin pattern tends to be disrupted by malignant but not by benign skin lesions suggests that measurements of skin pattern disruption on simply captured white light optical clinical (WLC) skin images could be a useful contribution to a diagnostic feature set. Previous work which generated a flow field of skin pattern using a measurement of local line direction and variation determined by the minimum eigenvalue and its corresponding eigenvector of the local tensor matrix to measure skin pattern disruption was computationally low cost and encouraging. This paper explores the possibility of extracting new features from the first and second differentiations of this flow field to enhance classification performance. METHODS: The skin pattern was extracted from WLC skin images by high-pass filtering. The skin line direction was estimated using a local image gradient matrix to produce a flow field of skin pattern. Divergence, curl, mean and Gaussian curvatures of this flow field were computed from the first and second differentiations of this flow field. The difference of these measures combined with skin line direction across the lesion image boundary was used as a lesion classifier. RESULTS: A set of images of malignant melanoma and benign naevi were analysed as above and the scatter plot in a two-dimensional dominant feature space using principal component analysis showed excellent separation of benign and malignant lesions. A receiver operating characteristic plot enclosed an area of 0.96. CONCLUSIONS: The experimental results show that the divergence, curl, mean and Gaussian curvatures of the flow field can increase lesion classifier accuracy. Combined with skin line direction they are promising features for distinguishing malignant melanoma from benign lesions and the methods used are computationally efficient which is important if their use is to be considered in clinical practice.     

Diagnostic image analysis of malignant melanoma in in vivo confocal laser-scanning microscopy: a preliminary study

Background/purpose: In this study we assessed the applicability of image analysis and a machine learning algorithm on diagnostic discrimination of benign and malignant melanocytic skin tumours in in vivo confocal laser-scanning microscopy (CLSM).
Methods: A total of 857 CLSM tumour images including 408 benign nevi and 449 melanoma images was evaluated. Image analysis was based on features of the wavelet transform. For classification purposes we used a classification tree software (CART). Moreover, automated image analysis results were compared with the prediction success of an independent human observer.
Results: CART analysis of the whole set of CLSM tumour images correctly classified 97.55% and 96.32% of melanoma and nevi images. In contrast, sensitivity and specificity of 85.52% and 80.15% could be reached by the human observer. When the image set was randomly divided into a learning (67% of the images) and a test set (33% of the images), overall 97.31% and 81.03% of the tumour images in the learning and test set could be classified correctly by the CART procedure.
Conclusion: Provided automated decisions can be used as a second opinion. This can be valuable in assisting diagnostic decisions in this new and exciting imaging technique.     

Skin lesion classification using relative color features

BACKGROUND/PURPOSE: Clinically, it is difficult to differentiate the early stage of malignant melanoma and certain benign skin lesions due to similarity in appearance. Our research uses image analysis of clinical skin images and relative color-based pattern recognition techniques to enhance the images and improve differentiation of these lesions. METHODS: First, the relative color images were created from digitized photographic images. Then, they were segmented into objects and morphologically filtered. Next, the relative color features were extracted from the objects to form two different feature spaces; a lesion feature space and an object feature space. The two feature spaces serve as two data models to be analyzed to determine the best features. Finally, we used a statistical analysis model of relative color features, which better classifies the various types of skin lesions. RESULTS/CONCLUSIONS: The best features found for differentiation of melanoma and benign skin lesions from this study are area, mean in the red and blue bands, standard deviation in the red and green bands, skewness in the green band, and entropy in the red band. With the relative color feature algorithm developed, the best results were obtained with a multi-layer perceptron neural network model. This showed an overall classification success of 79%, with 70% of the benign lesions successfully classified, and 86% of malignant melanoma successfully classified.     

Digital computer analysis of dermatoscopical images of 260 melanocytic skin lesions; perimeter/area ratio for the differentiation between malignant melanomas and melanocytic nevi

BACKGROUND: Digital computer analysis of dermatoscopical images has been reported to facilitate the differential diagnosis of pigmented skin lesions in recent years. OBJECTIVE: The aim of our study was to perform digital computer analysis of a set of different melanocytic lesions and compare the objective results. METHODS: The set of 260 melanocytic lesions (150 excised difficult cases (46 melanomas, 47 atypical nevi, 57 common nevi and 110 unexcised common nevi) was automatically analysed by the digital dermatoscopical system microDERM. We searched for differences in asymmetry, size, compactness and colour distribution. Perimeter/area ratio was calculated. RESULTS: The perimeter/area ratio was detected as the most important criterion for differentiation between malignant and benign melanocytic lesions (sensitivity 91.3% and specificity 90.7% for malignant melanomas vs. all benign nevi; sensitivity 91.3% and specificity 80.8% for melanomas vs. clinically atypical nevi). Differences in size of the lesion, shape and asymmetry of colour were found and statistically verified. Using step-wise logistic regression the formula for calculation of probability of malignant nature of every analysed lesion was constructed. CONCLUSION: The perimeter/area ratio is a simple parameter for the differential diagnosis of melanocytic skin lesions.     

Intraepidermal distribution patterns of melanocytes in the melanocytic lesions on the sole: proposed criteria for histopathologic diagnosis of plantar malignant melanoma in situ

Eighty-eight melanocytic lesions on the soles of Japanese were histologically investigated. Increased numbers of solitary melanocytes above the basal layer of the epidermis were often found in the benign melanocytic nevi on the sole: in 5 lesions of 9 congenital melanocytic nevi, 22 of 65 acquired melanocytic nevi, and 1 of 5 dysplastic nevi. In addition, a moderate degree of nuclear atypia of proliferating melanocytes was frequently observed in the benign melanocytic nevi on the sole: in 3 lesions of 9 congenital melanocytic nevi, 17 of 65 acquired melanocytic nevi, and 2 of 5 dysplastic nevi. Therefore it cannot be said that increased numbers of solitary atypical melanocytes above the basal layer is a characteristic histologic feature of early malignant melanoma in situ. Combining the intraepidermal distribution patterns of melanocytes and maximum diameter of the lesion, we propose criteria for histopathologic diagnosis of plantar malignant melanoma in situ.