Abstracts Submitted: 510
Number of Users: 710
View Abstracts Submitted
Back to home Page
Ultrasound has been widely used as a diagnostic tool in the field of medicine to visualize the internal body organs like kidneys, liver, breasts, thyroid etc. It has various advantages over other imaging modalities like cost-effectiveness, portability, versatile, and not harmful to the human beings. The only disadvantage of ultrasound is its operator/observer dependency. The heterogeneity of various abnormalities encountered in an organ, as visible on ultrasound leads to an overlap in the characterizing patterns, due to which the differential diagnosis between these abnormalities cannot be made clearly by the radiologists. Therefore, to reduce this ambiguity, an increasing amount of interest is encountered among the research community to develop different computer aided diagnostic (CAD) systems that can serve as efficient second opinion tools to overcome the inter-observer variation. These CAD systems accept ultrasound images as inputs and compute the texture and shape patterns of the abnormality under study in terms of some mathematical which are fed to different classification algorithms to diagnose the type of abnormality. In the present research work ultrasound images of the breast have been considered for analysis and classification of lesions. The work aims at designing a CAD system that considers both the characteristics of the lesions for the classification purposes using both machine learning and deep learning techniques. The ultrasound images may sometimes contain speckle noise, which affects the contrast of the image it becomes difficult for the radiologists to analyze the image clearly. Also due to the presence of speckle noise, some of the finer clinical details that might be helpful for detecting the abnormalities are masked. Therefore, the presence of speckle noise reduces ability of the radiologist to extract the vital information from a medical image visually for a concise diagnosis. Thus, in order to get a clear diagnosis about certain abnormality, methods have been employed to reduce the speckle noise without compromising any fine details from the ultrasound images. Initially for analysis purposes, 100 breast ultrasound images were used for the classification of lesions using raw images and their texture characteristics computed using Laws’ masks of various length. The study reported an accuracy of 80.39 % using SSVM classifier and PCA for feature dimensionality reduction.
© Copyright 2017 All Rights Reserved