Examining ultrasonic scattering mechanisms in fatty liver parenchyma via histopathological liver image

Wu, Yashuo

Permalink

https://hdl.handle.net/2142/113228 Copy

Description

Title

Examining ultrasonic scattering mechanisms in fatty liver parenchyma via histopathological liver image

Author(s)

Wu, Yashuo

Issue Date

2021-07-23

Director of Research (if dissertation) or Advisor (if thesis)

Han, Aiguo

Department of Study

Electrical & Computer Eng

Discipline

Electrical & Computer Engr

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

M.S.

Degree Level

Thesis

Keyword(s)

• Quantitative ultrasound
• Backscatter coefficient

Abstract

Quantitative ultrasound (QUS) techniques are diagnostically useful in assessing nonalcoholic fatty liver disease (NAFLD). From previous studies, the backscatter coefficient (BSC) correlates with liver fat fraction, but the mechanisms of the correlation have not been fully determined. Understanding the ultrasound scattering mechanisms will lead to better acoustic scattering models and potentially more accurate diagnostics. Several scattering mechanisms have been hypothesized and the following mechanism is tested in this thesis: Fat droplet deposition in hepatocytes alters the locations of hepatocyte nuclei, changing the spatial distribution of the nuclei, which leads to the change in the structure function, a factor of the BSC. The structure function, determined by the spatial distribution of the nuclei, is thus correlated with the fat fraction. Digitized histopathological liver slides were analyzed to test this hypothesis. Seventy-six (76) hematoxylin and eosin-stained liver histopathological slides from 46 participants with NAFLD and 10 participants without NAFLD were digitized at 40× objective magnification. One to five regions of interest (ROIs) (453.6 µm × 453.6 µm) were selected from each participant’s slide(s), yielding 258 ROIs in total. For each ROI, hepatocyte nuclei and fat droplets were automatically recognized using two independent methods: the morphological processing method and the U-Net method. Fat droplets were also automatically recognized for each ROI. In addition, hepatocyte nuclei were manually annotated for a subset of ROIs (N = 47). The structure function curves were calculated from the nuclear distributions obtained from the automatic and manual recognition methods. The liver fat fraction was determined from histology for each ROI by calculating the fractional surface area of fat droplets. Structure function was positively correlated with the liver fat fraction (Pearson’s r ~ 0.45, p < 10-4 ) below 25 MHz (including clinically relevant frequencies 3 - 5 MHz) using the morphological processing method for nucleus recognition. Pearson’s r value between the structure function and the fat fraction was approximately 0.40 (p < 10-4) below 20 MHz using the U-Net method for nucleus recognition. Structure function calculated from manually annotated nucleus positions had a positive correlation (Pearson’s r ~ 0.45, p < 0.01) with fat fraction below 25 MHz. In conclusion, this study shows the fat droplets change the distribution of hepatocyte nuclei, which is a factor contributing to the correlation between fat fraction and BSC.  

Graduation Semester

2021-08

Type of Resource

Thesis

Permalink

http://hdl.handle.net/2142/113228

Copyright and License Information

Copyright 2021 Yashuo Wu

Owning Collections