University of British Columbia
Automated volumetric ultrasound attenuation analysis for hepatic steatosis assessment
Abstract
dc:descriptionMetabolic dysfunction-associated steatotic liver disease (MASLD) affects more than 30% of adults worldwide, motivating scalable quantification of hepatic steatosis. Liver biopsy is invasive and subject to sampling variability, whereas MRI proton-density fat fraction (MRI-PDFF) provides a non-invasive imaging reference but is costly and not widely available. Ultrasound attenuation imaging is related to steatosis, but implementations depend on manually selected regions of interest, limiting tissue coverage and increasing operator dependence. This thesis developed and evaluated an automated post-acquisition framework for generating volumetric attenuation coefficient estimates (ACEs) from liver radiofrequency ultrasound data. The framework addressed three requirements for reliable automated ACE estimation in vivo: selecting representative liver parenchyma, identifying a reliable depth range for attenuation estimation, and integrating quality-controlled measurements. Learning-based tissue masking provided tissue-aware spatial filtering. Within the retained tissue, the attenuation-range detector identified axial intervals suitable for attenuation estimation. Per-line ACEs were estimated using the reference-phantom method, with model-based A-line quality control providing a final safeguard against unreliable fits. Accepted A-line estimates were aggregated to produce plane- and acquisition-level ACEs, and subsequently summarized at the participant level. Phantom and in vivo data were acquired using a Philips xMATRIX-based three-dimensional liver elastography system providing 22 planes per volume. Processing parameters were selected using phantom data and locked before evaluation on a held-out material. The fixed pipeline was then applied to 237 volumes from 35 participants with same-visit MRI-PDFF. Evaluation assessed phantom accuracy, association with MRI-PDFF, same-session repeatability, and discrimination at MRI-PDFF thresholds of 6%, 17%, and 22%. For the held-out phantom material, the pipeline produced an RMSE of 0.024 dB/cm/MHz. Participant-level median ACE was strongly associated with MRI-PDFF, with Pearson r = 0.8835 and Spearman ρ = 0.9135. Same-session repeatability yielded a mean within-subject coefficient of variation of 3.24% and a single-acquisition ICC(1, 1) of 0.902. The areas under the receiver operating characteristic curves were 0.958, 0.954, and 0.948 at the 6%, 17%, and 22% thresholds, respectively. These findings demonstrate the feasibility of automated volumetric ultrasound attenuation assessment. Further studies should assess robustness across imaging settings and ultrasound systems from different manufacturers.
Degree
thesis:*- Name thesis:degree_name
- Master of Applied Science - MASc
- Level thesis:degree_level
- master's
- Discipline thesis:degree_discipline
- Biomedical Engineering
- Grantor dc:publisher
- University of British Columbia
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Li, Zongze
- Contributors dc:contributor
-
- Salcudean, Septimiu Edmund
Rights
dc:rights- Statement dc:rights
-
- Attribution-NonCommercial-NoDerivatives 4.0 International
- Language dc:language
- eng
Identifiers
dc:identifier.*- Handle dc:identifier
- http://hdl.handle.net/2429/95210