Year: 2013
Type: article
Abstract: This work describes a semi-supervised approach to analyze the spectral data content of the samples of surface scanned skin lesion images for cancer diagnosis.As the first step, a surface image of the skin nevi is segmented utilizing Otsu's grayscale histogram thresholding technique.The anomalous part of the skin tissue is isolated from the surrounding normal skin cells appearing in the background of the skin sample image.A cost estimation function based on the mean and variance of localized windows extracts sub samples from the lesion that has the highest discriminative information.The coordinates of the LAB color channels of the segmented skin nevi are used to obtain the spectral information in each of the windows.The Euclidian distance is employed as the similarity measure tool to compare the color feature vectors of the samples of the three classes, benign, precancerous and malignant in the training and the testing phases of the pattern recognition system.Seventy five surface skin samples have been used for this distance based classification.Classifier performance metrics such as accuracy of 80%, sensitivity of 100% and specificity of 60% have been achieved during the implementation of this computer assisted clinical skin cancer diagnosis system.An adaptive homomorphic filter is designed to eliminate speckle noise in ultrasound scans and thereby increase the ability to perceive morphological characteristics of cells, tissue and organs.throughout the last three years.Heartfelt gratitude to my research group James Graham, John Willis, Melih Altun, Avishek Parajauli, Ivan Glasgo, and William Adams for always being there for me.Liz Masson, who has been
Author: Nikhil J. Dhinagar
Institution: Ohio University
Language: English
Subfield: Oncology
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W2616613480