Abstracts / IBRO Reports 6 (2019) S346–S562
S523
P35.08
into a certain area, thus concentrates antibody into the direction
of thick biological samples, enabling rapid, uniform, and complete
staining without sample distortion. Using the EFIC method, we
could stain thick brain tissues uniformly and rapidly (up to 3 mm
deep sample within 4 h) with only a limited amount of antibody
(typically 50 g/reaction). We have successfully applied EFIC to
formalin-fixed postmortem human brain tissues and simultaneous
multi-color staining with 4 different antibodies. We also developed
a novel optical clearing solution named perfectMATCH to minimize
deformation artifacts that are due to the harsh treatment and tran-
sient sample swelling. Together, our EFIC and perfectMATCH are
optimized to preserve all necessary 3D molecular and structural
information for high-resolution fluorescence imaging.
Pediatric brain extraction from T2-weighted MR
images using 3D dual frame U-net and human
connectome database
Dongchan Kim1, Jong-Hee Chae2, Sunkyue
Kim3, Yeji Han1,∗
1 Gachon University, Incheon, Republic of Korea
2 Seoul National University College of Medicine,
Seoul, Republic of Korea
3 Neuroscience Research Institute, Gachon
University, Incheon, Republic of Korea
Objective: The purpose of this study was to investigate a brain
extraction technique for multi-slice 2D pediatric MR images with
disease using the HCP database and dual frame (DF) 3D U-net deep
learning architecture.
P35.10
Materials and methods: To compensate for the limited data of
MR images and manual segmentation masks of pediatric patients,
we pre-trained networks with the HCP database. DF 3D U-net archi-
tecture was used to minimize segmentation errors in the small
object and boundary regions. We quantitatively analyzed the per-
formance of the BET, DF, and conventional 3D U-net for brain
extraction using the IoU and BF scores; each deep learning architec-
ture was evaluated with and without pre-training using the HCP.
This study included 10 patients with diseases and all images were
acquired using a TSE sequence with a BLADE trajectory.
Results: We showed that pre-training using the HCP database
could enhance segmentation performance of the network, and that
skip connections in the DF 3D U-net could enhance the contour sim-
ilarity of segmentation results. Experimental results showed that
the proposed method increased values of IoU and BF score by 0.002
and 0.003, respectively, compared with those of the conventional
3D U-net without pre-training.
Miniaturized ultrasound systems for
modulation of sleep of freely moving mice
Yehhyun Jo, Sang-Mok Lee, Seongyeon Kim,
Hyunggug Kim, Hyunjoo Lee∗
KAIST, Daejeon, Republic of Korea
Low-intensity transcranial Focused Ultrasound Stimulation
(tFUS) is a promising new modality that offers distinct advantages
such as noninvasiveness, high spatial resolution, deep penetration
depth, and the ability to conduct clinical trials. However, despite the
recent efforts into developing miniaturized systems capable of tFUS
on freely-moving small animals, there have been no reported physi-
ological or behavioral studies employing a miniaturized transducer
system. Thus, we developed a novel capacitive micromachined
ultrasonic transducer (CMUT) ring array device, real-time detec-
tion and sonication system, and custom sleep analysis program. By
using this system, we stimulated the prefrontal cortex (PFC) dur-
ing non-rapid eye movement (NREM) sleep in freely-moving mice.
Our experiment cycle consisted of four consecutive days. On the
first day we set the baseline sleep stage durations, on the second
day we delivered tFUS, and on the following two days we con-
ducted observations. Our experimental results show both acute and
chronic increases in the NREM and REM sleep durations. On the day
of stimulation, we observed a significant increase in the total REM
duration in the stimulated mice, a dramatic decrease in the NREM
duration in the control/sham mice, and an unchanged NREM dura-
tion in the stimulated mice. On the third and fourth observational
days, we observed notable increases in the REM and NREM sleep
durations during certain 1-h intervals of sleep in the stimulated
mice. Here, we demonstrated for the first time that tFUS of the PFC
during NREM sleep in freely-moving mice both acutely and chroni-
cally modulated the sleep patterns of behaving small animals. This
study on the effect of tFUS on sleep presents ultrasound stimulation
as a promising new therapeutic means for sleep disorders.
Conclusion: We showed that the DF 3D U-net with pre-training
using the HCP database could enhance the quality of brain extrac-
tion results. More specifically, we demonstrated that utilizing the
HCP database for pre-training in human brain semantic segmenta-
tion tasks, which comprised brain extraction in our experiments,
could compensate for an insufficient amount of training data.
P35.09
Rapid and uniform staining of thick biological
tissues with antibody using Electro-magnetic
Focused Immuno-histoChemistry
Myeongsu Na, Kitae Kim, Sunghoe Chang∗
Seoul National University College of Medicine, Seoul,
Republic of Korea
Understanding the structure of a complex biological system
at the cellular level requires 3D anatomical and phenotypical
maps. This task has dramatically advanced by recent surges of
various tissue clearing methods. Labeling of thick tissue with
an antibody, however, still depends on the slow diffusional
process, thus suffer from inefficient and uneven penetration of
macromolecules into thick tissues. Although a few methods have
developed to overcome this problem, they are still laborious and
require specialized equipment. Here, we present a novel technique
for rapid and uniform penetration of antibody into thick biological
tissues using the focused electro-magnetic forces. This technique
that we called the EFIC (Electro-magnetic Focused Immuno-
histoChemistry) achieves the focused electric and magnetic field
Kim, In-Hae
