Figure 3. Plots and linear regressions of triglyceride content vs.
converted fat contents from voltage signals of the reflectance and Raman
channels. Shadow areas: 95% confidence bands of linear regressions.
An expert liver pathologist (AES) assessed the whole slide images (as
shown in Figure 4A ) and estimated, to the nearest 5% interval,
the percent of large droplet MaS and the percent of global (macro- and
micro-vesicular) HS. The assessment of large droplet MaS was performed
per the most recent Banff recommendations12 for
“large droplet fat,” defined as a single droplet expanding the
hepatocyte and larger than adjacent cells, with low power determination
of the percentage of tissue affected, followed by higher magnification
assessment of fatty areas to determine what proportion of steatotic
cells meet the definition of “large droplet fat,” and finally
adjustment of the low power percentage accordingly. After completion of
pathologist estimates, positive pixels (as shown in Figure 4B )
were counted automatically using the Positive Pixel Count algorithm v9
of Aperio ImageScope (Leica Biosystems); this value was normalized by
the area assessed to generate a percentage. Areas of artifact, such as
tissue folding or coverslip lifting, were manually excluded from the
analysis.