Needle Detection And Localization In Simulated In-vivo Ultrasound Image For Use In Breast Biopsy
Abstract
This thesis addresses needle segmentation in ultrasound images for breast cancer biopsy. The thesis reviews the state-of-the-art in object localization using ultrasound (US) imagery. In particular, principal component analysis, projection based methods (Hough Transform, Radon Transform and Parallel Integral Projection) and model fitting method using RANSAC and work in the field of diffusion filters are reviewed for potential use in a real breast biopsy environment, wherein the US imagery does not have much contrast and the speckle noise content is high. Other image artifacts in these images also mask the needle image.A new needle localization method called ANDSUI algorithm has been developed with focus on improved localization accuracy and computational complexity especially when the assumption "needle is brighter than background" fails such, in cases like needle is behind dense tissue. The new method and the state of the art methods have been compared to show that ANDSUI algorithm is superior.