Discussion
The accumulation of TGs within hepatocytes is characteristic of
HS.35 Raman spectroscopy has been validated to assess
liver steatosis in animal models based on characteristic Raman
peaks.20,21 While the calibration of the investigated
Raman system was in keeping with previous reports,26the investigated Raman system demonstrated substantial positive
correlations with both Raman (r = 0.82) and reflectance (r = 0.64)
channels, indicating its capability to assess TG in liver specimens.
This made it suitable for further evaluation in the histopathological
validation stage.
In the histopathological validation stage, the Raman channel was
revealed to be able to classify MaS and global HS of human liver
specimens. Combining information from the two channels, dual-variable
predictions of 66 liver specimens reported good (AUROC = 0.80)
classification of an ≥10% steatosis discrepancy between
pathologist-estimated MaS and the algorithm-estimated HS (ORO-stained
pixel area). These findings indicated that the investigated ambient
light-compatible Raman system was capable of accurate ex-situ evaluation
of global HS in human livers. The investigated Raman system has
demonstrated preliminary efficacy in MaS assessments, indicating its
potential as a valuable tool for the in-situ real-time surgical
evaluation of donor liver steatosis.
The primary strength of the present study lies in the multifaceted
correlation of Raman assessments with a robust array of subjective and
objective references. This included TG quantification, expert
pathologist evaluation, and advanced the algorithm-based estimation,
achieving validations from diverse and complementary perspectives.
Moreover, the observed steatosis within the specimens in the
histopathological validation stage exhibited a broad distribution of
minimum-to-severe steatosis, facilitating a comprehensive evaluation of
the HS classifications through the investigated Raman system. This
wide-ranging analysis underscores the potential of our approach in
advancing the understanding of integrated spectroscopic evaluation on HS
and may pave the way for surgical diagnostic techniques.
All specimens included in this study were initially snap-frozen.
Although the specimens could not represent donor livers for recovery,
their chemical compositions (to which Raman spectroscopy is sensitive)
and morphology were to a great extent comparable to those of donor liver
organs. All the specimens studied in the pathological validation stage
were several times thicker than the 1-mm laser penetration depth
reported in the previous clinical trial,26 enabling
enough scattering interaction as a shone liver organ would provide.
This study encountered several limitations in evaluating the Raman
system for HS assessment. A significant hurdle was the mixed presence of
MaS and MiS in most specimens, complicating the distinction between MaS
and global HS. During the biochemical validation stage, the exclusion of
9 out of 25 specimens due to their size and weight markedly diminished
the statistical power, and the reliance on duck fat-agar phantoms for
setting up a “zero point” by calibration hampered the system’s
accuracy and convenience.
The research noted a promising dual-variable prediction based on Raman
scattering and reflectance, capable of classifying specimens with a
minimum 10% greater global HS than MaS. However, this finding, lacking
a solid physicochemical basis, necessitates further exploration. van
Staveren et al. (1991) accurately predicted the scattering coefficient
of fat emulsion lntralipid-10% using Mie theory and the size
distribution of lipid particles.36 A reasonable guess
is that the reflectance channel (Mie scattering) contained information
about both fat content and lipid droplet size; however, more studies
according to the physical principles of Raman and Mie scattering, e.g.,
Monte Carlo simulations, are needed to probe the concept of the
dual-variable prediction. Additionally, the untested performance of the
Raman system in operating room conditions, despite its proven resistance
to 10,000 lm LED light in laboratory settings, leaves a significant gap
in understanding its comprehensive compatibility.26In this study, a single expert pathologist (AES) assessed all
microscopic slide images and estimated the percentage of large droplet
MaS and global HS. Despite the potential for diverse insights from
multiple pathologists,37,38 we opted for a single
expert to ensure consistent assessment methods and mitigate the risk of
interobserver discordance. This approach also assured the quality of the
assessment, leveraging the pathologist’s high expertise. The Positive
Pixel Count algorithm, as an additional independent approach to quantify
the fat content, validated the pathologist’s estimations and enabled
finer quantitive discrepancy of global HS and MaS.
To the best of the authors’ knowledge, the present study represents the
inaugural trial of conventional Raman spectroscopy on human liver
specimens for HS assessment. Advancing beyond the scope of preceding
Raman studies, this research emphasized an ambient light-compatible
methodology. The efficacy of the Raman spectroscopic approach in
assessing the HS of human liver specimens was validated by correlating
the fat content results provided by the examined Raman system with TG
quantifications. Additionally, histopathological validation was
conducted across a wide range of steatotic human liver specimens,
culminating in the development of a dual-variable prediction for
significant discrepancies (>10%) between global HS and MaS.
The insights gained from this study may contribute to enhancing other
spectroscopic surgical instruments.