Noise reduction for ultrasound images using deep interpolation

Ultrasound imaging has been proven to be a safe and effective method of detecting signals in the body related to physiological parameters such as anatomy, blood flow, and stiffness, aiding in the diagnosis of various diseases. However, the quality of these images can be compromised by the low signal-to-noise ratio (SNR) caused by electronic interference noise. Traditional methods like frame averaging can increase SNR, but they require a large number of input frames to produce a single frame of high-SNR output ultrasound images and are ineffective when motion is present as the output image might be blurry. Deep-learning-based denoising techniques have been developed throughout the years, but they either require ground truth images or identical signals in all time frames, which are unachievable in in-vivo ultrasound imaging. To address these limitations, we utilized a U Net-based encoder-decoder network called Deep Interpolation, which uses 40 input frames to reconstruct a high-quality noise-reduced output ultrasound image without ground truth and can be used for both static and in-vivo ultrasound images.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Electrical & Computer Engr

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2023

Author dc:creator

• Li, Xiaobai

Contributors dc:contributor

• Song, Pengfei

• Ultrasound
• Noise Reduction
• Deep Interpolation

Statement dc:rights

• Copyright 2023 Xiaobai Li

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/120586

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F120586