Titanium bead calibration of deep net classifiers

With the rapid development of computational power in the last decade, the field of deep learning and its applications have advanced greatly. The field of medical ultrasound specifically has greatly benefited from the integration of deep learning; however, it is often hindered by domain differences between training and deployment. In order to solve this issue, references phantoms have been used in the past to calibrate domains; however, those are unable to calibrate differences within the tissue especially when large differences exist between training and testing domains due to tissue attenuation. In this work, we examine the use of an in situ titanium bead and its potential use as a calibration signal to allow deep learning models to generalise between training and testing domains. It is determined that the calibration using this bead can lead to improvements in classifier accuracy from 50\% up to 93\%.

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

2024

Author dc:creator

• Cai, William

Contributors dc:contributor

• Oelze, Michael L

• Ultrasound
• Machine Learning
• Deep Learning
• Domain Adaptation

Statement dc:rights

• Copyright 2024 William Cai

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/124470

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F124470