Open AccessFixed grid wavelet network segmentation on diffuse optical tomography image to detect sarcoma
Author(s) -
K. Uma Maheswari,
S. Sathiyamoorthy
Publication year - 2018
Publication title -
journal of applied research and technology
Language(s) - English
Resource type - Journals
ISSN - 2448-6736
DOI - 10.22201/icat.16656423.2018.16.2.706
Subject(s) - artificial intelligence , wavelet , grayscale , mean squared error , mathematics , segmentation , pattern recognition (psychology) , pearson product moment correlation coefficient , computer science , computer vision , image (mathematics) , statistics
Objective: To detect and explore the boundary of the sarcoma in Diffuse Optical Tomography (DOT) images, we need to extract the scattering and absorption property of the tissue at the cellular level. The DOT images suffer with lower optical resolution; therefore to improve the resolution in non-invasive imaging technique we apply Fixed Grid Wavelet Network (FGWN) image segmentation. Methods: We have subjected the reconstructed optical image to Vignette Correction to enhance the corners so that it traces the smooth boundary of tumor region. Fixed Grid Wavelet Network segmentation applied to reduce the training with the significant ortho-normal property. R, G and B values of optical image were considered as network inputs which lead to the formation of Wavelet network. Effective wavelet selection was based on Orthogonal Least Squares Algorithm and the network weights were calculated to optimize the network structure. The Mexican hat wavelet chosen facilitates the diffusion operator for image restoration, hence well-suited for Diffuse Optical Tomography (DOT) images.Results: Analysis made on data base of 30 DOT images and the 6 criteria results was evaluated. The boundary of the tumor region was traced on grayscale and the following Image Metrics were measured namely Mean Square Error, Root Mean Square Error, Peak Signal to Noise Ratio, Pearson Correlation Coefficient and Mean absolute error. The Receiver Operating Characteristics (ROC) was estimated at 99.527%, 88.73% and 93.8% with respect to sensitivity, specificity and overall accuracy. Conclusions: FGWN was compared with genetic algorithm and graph cut segmentation based on image metrics which exhibited 5.2% improvement and it was evaluated such that FGWN based image segmentation was superior to other methodologies.