BEGIN:VCALENDAR
VERSION:2.0
PRODID:IEEE vTools.Events//EN
CALSCALE:GREGORIAN
BEGIN:VTIMEZONE
TZID:America/Chicago
BEGIN:DAYLIGHT
DTSTART:20200308T030000
TZOFFSETFROM:-0600
TZOFFSETTO:-0500
RRULE:FREQ=YEARLY;BYDAY=2SU;BYMONTH=3
TZNAME:CDT
END:DAYLIGHT
BEGIN:STANDARD
DTSTART:20191103T010000
TZOFFSETFROM:-0500
TZOFFSETTO:-0600
RRULE:FREQ=YEARLY;BYDAY=1SU;BYMONTH=11
TZNAME:CST
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20200122T185417Z
UID:30F7C2A3-8176-4215-9F91-06BD9503E52D
DTSTART;TZID=America/Chicago:20200120T183000
DTEND;TZID=America/Chicago:20200120T200000
DESCRIPTION:An automated and generally applicable method for segmentation i
 s still a focus of medical image processing research. Artificial inteligen
 ce methods have shown promising results in recent years\, especially with 
 widely available scalable Deep Learning libraries. In this work\, a five l
 ayer hybrid U-net is developed for slice-by-slice segmentation of liver da
 ta sets. Training data is taken from the Medical Segmentation Decathlon da
 tabase\, providing 131 fully segmented volumes. A slice-oriented segmentat
 ion model is implemented utilizing deep learning algorithms with adaptions
  for variable parenchyma shape along the stacking direction and similariti
 es between adjacent slices. Both are transformed for coronal and sagittal 
 views. The implementation is on a GPU rack with TensorFlow and Keras. For 
 a quantitative measure of segmentation accuracy\, standardized volume and 
 surface metrics are used. Results DSC=97.59\, JI=95.29 and NSD=99.37 show 
 proper segmentation comparable to 3D U-Nets and other state of the art. Th
 e development of a 2D-slice oriented segmentation is justified by short tr
 aining time and less complexity and therefore massively reduced memory con
 sumption.\n\nWe further evaluated how the U-net cascade can be applied to 
 different medical domains with less available reference data for training\
 , more precisely MRI data of the liver from VIBE acquisition protocol. Wit
 h Graph cuts as a post-processing tool\, expert knowledge can overcome the
  black-box nature of Deep Learning models in a semi-automated way. Weights
  for combining original image edges and edges from the DL segmentations re
 sult from Evolution strategy optimization\, closing the gap in accuracy of
  Deep Learning and Graph cut in general but still conserving the potential
  for expert-driven corrections. Although the U-net cascade lacks a suffici
 ent amount of MRI training data leading to mediocre accuracy\, Graph cuts 
 is a powerful tool for post-processing and iterative enrichment of the tra
 ining basement. Another strategy for enriching small reference datasets ar
 e generative adversial networks (GANs). It is shown that datasets enriched
  by synthesized slices improve the training process of Deep Learning model
 s.\n\nCo-sponsored by: Victoria Marks\n\nSpeaker(s): Gerald Zwettler\, Ph.
 D.\, \n\nAgenda: \nA pizza social will begin at 6:30 PM\, followed by the 
 start of the talk at 7:00 PM.\n\nRoom: Mann Hall\, Bldg: Medical Sciences 
 Bldg (east side\, north door)\, 321 3rd Ave SW\, Rochester\, Minnesota\, U
 nited States\, 55902
LOCATION:Room: Mann Hall\, Bldg: Medical Sciences Bldg (east side\, north d
 oor)\, 321 3rd Ave SW\, Rochester\, Minnesota\, United States\, 55902
ORGANIZER:v.marks@ieee.org
SEQUENCE:2
SUMMARY:Pizza and Lecture: U-Net Deep Learning Strategies for Slice-wise Se
 gmentation of Tomographic 3D Datasets
URL;VALUE=URI:https://events.vtools.ieee.org/m/217912
X-ALT-DESC:Description: <br /><p>An automated and generally applicable meth
 od for segmentation is still a focus of medical image processing research.
  Artificial inteligence methods have shown promising results in recent yea
 rs\, especially with widely available scalable Deep Learning libraries.&nb
 sp\;In this work\, a five layer hybrid U-net is developed for slice-by-sli
 ce segmentation of liver data sets. Training data is taken from the Medica
 l Segmentation Decathlon database\, providing 131 fully segmented volumes.
  A slice-oriented segmentation model is implemented utilizing deep learnin
 g algorithms with adaptions for variable parenchyma shape along the stacki
 ng direction and similarities between adjacent slices. Both are transforme
 d for coronal and sagittal views. The implementation is on a GPU rack with
  TensorFlow and Keras. For a quantitative measure of segmentation accuracy
 \, standardized volume and surface metrics are used. Results&nbsp\;<em>DSC
 =97.59</em>\,&nbsp\;<em>JI=95.29</em>&nbsp\;and&nbsp\;<em>NSD=99.37</em>&n
 bsp\;show proper segmentation comparable to 3D U-Nets and other state of t
 he art. The development of a 2D-slice oriented segmentation is justified b
 y short training time and less complexity and therefore massively reduced 
 memory consumption.</p>\n<p>We further evaluated how the U-net cascade can
  be applied to different medical domains with less available reference dat
 a for training\, more precisely MRI data of the liver from VIBE acquisitio
 n protocol. With Graph cuts as a post-processing tool\, expert knowledge c
 an overcome the black-box nature of Deep Learning models in a semi-automat
 ed way. Weights for combining original image edges and edges from the DL s
 egmentations result from Evolution strategy optimization\, closing the gap
  in accuracy of Deep Learning and Graph cut in general but still conservin
 g the potential for expert-driven corrections. Although the U-net cascade 
 lacks a sufficient amount of MRI training data leading to mediocre accurac
 y\, Graph cuts is a powerful tool for post-processing and iterative enrich
 ment of the training basement. Another strategy for enriching small refere
 nce datasets are generative adversial networks (GANs). It is shown that da
 tasets enriched by synthesized slices improve the training process of Deep
  Learning models.</p><br /><br />Agenda: <br /><p>A pizza social will begi
 n at 6:30 PM\, followed by the start of the talk at 7:00 PM.</p>
END:VEVENT
END:VCALENDAR