Melanoma Identification Using MLP With Parameter Selected by Metaheuristic Algorithms

Melanoma Identification Using MLP With Parameter Selected by Metaheuristic Algorithms

Soumen Mukherjee (RCC Institute of Information Technology, India), Arunabha Adhikari (West Bengal State University, India) and Madhusudan Roy (Saha Institute of Nuclear Physics, India)
DOI: 10.4018/978-1-5225-7107-0.ch010
OnDemand PDF Download:
No Current Special Offers


Nature-inspired metaheuristic algorithms find near optimum solutions in a fast and efficient manner when used in a complex problem like finding optimum number of neurons in hidden layers of a multi-layer perceptron (MLP). In this chapter, a classification work is discussed of malignant melanoma, which is a type of lethal skin cancer. The classification accuracy is more than 91% with visually imperceptible features using MLP. The results found are comparably better than the related work found in the literature. Finally, the performance of two metaheuristic algorithms (i.e., particle swarm optimization [PSO] and simulated annealing [SA]) are compared and analyzed with different parameters to show their searching nature in the two-dimensional search space of hidden layer neurons.
Chapter Preview


Automatic detection of malignant Melanoma is carried out by several researchers starting from the last decade. In 2015 Giotis (Giotis et al. 2015) used only 675 features to attain 81% of accuracy with MED-NODE dataset of 170 images. After pre-processing all the images are converted to 50 square patches of 15 X 15 pixels size which finally classified by CLAM classifier. In the year 2016 Tan (Tan et al. 2016) reached to 88% accuracy rate using Dermofit image library of 1300 images using SVM classifier. Preprocessing steps like removal of hair, enhancement of contrast are done in each image. Genetic Algorithm used as feature selector to select 1472 features out of 3914 features extracted. Laskaris (Laskaris et al., 2010) and McDonagh (McDonagh et al., 2008) worked on Dermofit dataset and achieved accuracy of 80.64% with only 31 images and accuracy of 83.7% with 234 images respectively. An online Melanoma detection system is developed by Iyatomi (Iyatomi et al., 2005, 2006) using neural network with a classification accuracy of 97.3% with dermoscopy image unlike the present work. Esteva et al. (Esteva et al., 2017) recently used Deep Convolutional Neural Networks (CNN) for skin cancer classification. They have used 129450 number of image dataset in their work. Codella et al. (Codella et al., 2017) achieved 76% accuracy using deep learning with standard dataset.

Key Terms in this Chapter

Particle Swarm Optimization: A popular natured inspired metaheuristic technique modeled upon the actions of animals.

Hidden Layer Neuron: Number of neuron in the hidden layer of the multi-layer perceptron.

Malignant Melanoma: A type of skin cancer causes due to excessive UV ray exposure in skin melanocytes cell.

GLRLM: Gray level run length matrix is used to find pixel intensity run length texture feature in an image.

GLCM: Gray level co-occurrence matrix is used to find pixel intensity co-occurrence texture feature in an image.

Otsu Algorithm: An algorithm to finds the threshold intensity value which maximize inter class variance and minimize the intra class variance.

Simulated Annealing: A popular single solution metaheuristic algorithm which has a background of statistical mechanics.

Complete Chapter List

Search this Book: