Integration of modified ABCD features and support vector machine for skin lesion types classification
Integration of modified ABCD features and support vector machine for skin lesion types classification
复制标题
DOI：
10.1007/s11042-020-10056-8
复制
发表时间：
2020-11-05
影响因子：
3.6
通讯作者：
Raj, Y. Jacob Vetha
中科院分区：
文献类型：
--
作者：
Melbin, K.;Raj, Y. Jacob Vetha
由于皮肤细胞暴露在阳光下，皮肤细胞的异常生长常常会导致皮肤癌。皮肤病主要由细菌、真菌、病毒、紫外线和化学物质引起。一般来说，由于黑色素瘤、脂溢性角化病和红斑狼疮疾病的特征与色素性疾病的相似性，临床医生对它们进行分类一直是个难题。本文提出了一种从皮肤镜图像中检测皮肤病变的综合方法。所提出的基于改进的ABCD特征和支持向量机（SVM）的集成累积水平差均值（CLDM）已用于皮肤病变图像的检测和分类。所提出的修改后的 ABCD 特征用于从皮肤病变图像中提取形状、大小、颜色和纹理等皮肤特征。在分类方法之前，使用特征向量中心性特征排序和选择（ECFS）方法来实现更好的分类。经过特征选择方法后，通过支持向量机（SVM）对皮肤病变图像进行分类。通过评估 Jaccard 相似性指数 (JSI)、Dice 相似性系数 (DSC)、敏感性、特异性和准确性来评估分割性能。所提出的 SVM 方法对三种皮肤病变类别进行分类，并产生了优异的分类结果，对于黑色素瘤、脂溢性角化病和红斑狼疮，分类准确度为 97%，特异性为 98%，敏感性为 97%，JSI 为 97%，DSC 为 98%。所提出的方法以高精度对三种皮肤病变类别（黑色素瘤、脂溢性角化病和红斑狼疮）进行分类。集成方法不仅提高了准确性水平，而且还为更好的分类提供了重要信息。
The abnormal growth of skin cells often leads to skin cancer due to the exposure of skin cells to the sun. The skin disease is primarily caused by bacteria, fungus, viruses, UV radiation, and chemical substances. Generally, clinicians have been a trouble to categorize melanoma, seborrheic keratosis and lupus Erythematosus diseases due to the resemblance in the features of pigmented diseases. The paper presents an integrated approach for detecting the skin lesion from the dermoscopic images. The proposed integrated cumulative level difference mean (CLDM) based modified ABCD features and Support vector machine (SVM) have used for the detection and classification of skin lesion images. The proposed modified ABCD features employed for extracting the skin features like shape, size, color, and texture from the skin lesion images. Prior to the classification method, the Eigenvector Centrality feature ranking and selection (ECFS) method has utilized for better classification. After the feature selection method, a skin lesion image is classified by the Support vector machine (SVM). The performance of segmentation has been assessed by evaluating the Jaccard Similarity Index (JSI), Dice similarity coefficient (DSC), sensitivity, specificity, and accuracy. The proposed SVM method classifies the three skin lesion classes and produces excellent classification results with the classification accuracy is 97%, specificity is 98%, sensitivity is 97%, JSI is 97% and DSC is 98% for melanoma, seborrheic keratosis, and lupus Erythematosus respectively. The proposed approach classifies the three skin lesion classes (melanoma, seborrheic keratosis and lupus Erythematosus) with high accuracy. The integrated method not only enhances the accuracy level but also delivers significant information for better classification.