FIGURE 1. Proposed thrombus composition prediction compared to
the current practice. Conventionally thrombus composition is manually
analyzed though stained bright-field (BF) images. Our framework fusing
ODT and DL provides label-free and fully-automated prediction of the
composition. Scale bar = 500 µm.
This work exploits the high-resolution and quantitative 3D imaging
capability of ODT, which extract various quantitative biophysical
properties from the RI distribution.[10] These
strengths of label-free imaging and ODT had been demonstrated in
histological studies [12-17], including kidney
cancer,[18] kidney
injury,[19] and intestinal
inflammation.[20] The use of refractive index as
reproducible and quantitative imaging contrast has strengths when
combined with DL. Recently, various cellular and subcellular RI
distributions have been analyzed using the combination of HT and DL,
including cellular segmentation,[21-23] the
detection of biological compartments,[24-26] and
domain translations.[27,28]
Our approach facilitates label-free and automated examination of an AIS
thrombus by utilizing DL to predict the composition of the thrombus from
a 3D RI tomogram. We validated the prediction from each of the three
unstained thrombus slices by comparing to a successive slice that is
Hematoxylin and Eosin (H&E) stained. An accuracy of 95% was achieved
in the patch-wise classification when compared to board-certified
pathologists’ annotation. Furthermore, the spatial distribution of
thrombus composition was in high agreement with the annotation over the
entire slide, highlighting the scope for whole-slide analyses. We expect
that our study will enable rapid and quantitative evaluation of thrombus
composition, aiding clinicians in treating AIS efficiently.