Speckle Detection in Ultrasonic Images Using Unsupervised Clustering Techniques

Research for the improvement of the quality of clinical ultrasound images has been a topic of interest for researchers and physicians. One of the challenges is the presence of speckle artifacts. This dissertation reviews the speckle phenomena in such images, and develops algorithms to better identify this artifact in sonographic images. Speckle artifact is categorized into two groups: partially developed speckles and fully developed speckles (FDS). This concept has been used, along with the classification techniques, to segment the ultrasound images into patches and classify the patches in the image as FDS or non-FDS. The proposed algorithms and the results of the experiments have been validated using simulation, phantom and real data that were created for the purposes of this study or taken from other research groups. Current speckle detection methods do not optimize statistical features and they are not based on machine learning techniques. For the first time this work introduces a novel method for searching and extracting the best features for optimizing speckle detection rate using statistical machine learning and ensemble classification. Potential applications include strain imaging by tracking speckle displacement, elastography, speckle tracking and suppression applications, and needle-tracking applications.Ph.D., Biomedical Engineering -- Drexel University, 201