Computer-Aided Diagnosis of Malignant Melanoma Using Gabor-Based Entropic Features and Multilevel Neural Networks

Abstract

The American Cancer Society has recently stated that malignant melanoma is the most serious type of skin cancer, and it is almost 100% curable, if it is detected and treated early. In this paper, we present a fully automated neural framework for real-time melanoma detection, where a low-dimensional, computationally inexpensive but highly discriminative descriptor for skin lesions is derived from local patterns of Gabor-based entropic features. The input skin image is first preprocessed by filtering and histogram equalization to reduce noise and enhance image quality. An automatic thresholding by the optimized formula of Otsu’s method is used for segmenting out lesion regions from the surrounding healthy skin regions. Then, an extensive set of optimized Gabor-based features is computed to characterize segmented skin lesions. Finally, the normalized features are fed into a trained Multilevel Neural Network to classify each pigmented skin lesion in a given dermoscopic image as benign or melanoma. The proposed detection methodology is successfully tested and validated on the public PH2 benchmark dataset using 5-cross-validation, achieving 97.5%, 100% and 96.87% in terms of accuracy, sensitivity and specificity, respectively, which demonstrate competitive performance compared with several recent state-of-the-art methods.

A Machine Vision Approach for Classification of Skin Cancer Using Hybrid Texture Features

“…It was examined that 18.1 million people have been diagnosed with skin cancer every year [ 1 ]. The specific area of the skin in the human body is changed due to skin tissue damage [ 2 ]. Skin color and texture change during skin infection.…”

A Framework for Instantaneous Driver Drowsiness Detection Based on Improved HOG Features and Naïve Bayesian Classification

“…For light compensation, an adaptive contrast-limited histogram equalization algorithm [ 25 , 26 ] is then applied, with which each color channel is independently equalized to produce a better lighting-compensated image which further serves as input to the subsequent face-detection module. After the light compensation, the resolution of the light-compensated image can be reduced to increase the computational efficiency of the framework [ 27 ].…”

Towards Accurate Diagnosis of Skin Lesions Using Feedforward Back Propagation Neural Networks

“…An accuracy of 93.7% for melanoma detection was reported for a total number of 206 images (119 melanoma and 87 non-melanoma type). In addition, computational tools to extract and learn high-level features automatically from raw images such as Convolutional Neural Networks (CNNs), fully convolutional residual network (FCRN), Google’s Inception v4 CNN, VGGNet Convolution Neural Network or deep residual networks (ResNets) have drawn researchers’ attention in recent years for melanoma detection and classification [ 34 , 35 , 36 , 37 ]. Kaymak et al [ 36 ] used four FCNs, i.e., FCN-AlexNet, FCN-8s, FCN-16s and FCN-32s, in order to segment the skin lesions in images belonging to the ISIC 2017 dataset.…”

“…The performance accuracy ranged from 0.932 to 0.939 and the elapsed time was from 176 min (for FCN-AlexNet) to 508 min (for FCN-32s). Bakheet and Al-Hamadi [ 37 ] proposed a fully automated ANN for real-time melanoma detection using Gabor-based entropic features as highly discriminative descriptors for skin lesions. A Multilevel Neural Network (MNN) with an improved backpropagation algorithm provided an accuracy of 97.50%, sensitivity of 100% and specificity of 96.87% when 200 8-bit RGB dermoscopic images of melanocytic lesions from the PH2 database were analyzed.…”