Ultrasound imaging operation capture and image analysis for speckle noise reduction and detection of shadows

Alruhaymi, Maryam (2018) Ultrasound imaging operation capture and image analysis for speckle noise reduction and detection of shadows. Masters thesis, Memorial University of Newfoundland.

[English] PDF - Accepted Version Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission. Download (8MB)

Abstract

Ultrasound is becoming increasingly important in medicine, both as a diagnostic tool and as a therapeutic modality. At present, experienced sonographers observe trainees as they generate hundreds of images, constantly providing them feedback and eventually deciding if they have the appropriate skills and knowledge to perform ultrasound independently. This research seeks to advance towards developing an automated system capable of assessing the motion of an ultrasound transducer and differentiate between a novice, an intermediate and an expert sonographer. The research in this thesis synchronizes the ultrasound images with three depth sensors (Microsoft Kinect) placed on the top, left and right side of the patient to ensure the visibility of the ultrasound probe. Videos obtained from the three categories of sonographers are manually labeled and compared using Studiocode Development Environment to complete the items on the medical form checklist. Next, this thesis investigates and applies well known techniques used to smooth and suppress speckle noise in ultrasound images by using quality metrics to test their performance and show the benefits each one can contribute. Finally, this thesis investigates the problem of shadow detection in ultrasound imaging and proposes to detect shadows automatically with an ultrasound confidence map using a random walks algorithm. The results show that the proposed algorithm achieves an accuracy of automatic detection of up to 85%, based on both the expert and manual segmentation.

Actions (login required)

View Item