Detail publikačního výsledku

UTILIZATION OF CONVOLUTIONAL NEURAL NETWORKS FOR SEGMENTATION OF MOUSE EMBRYOS CARTILAGINOUS TISSUE IN MICRO-CTDATA

POLÁKOVÁ, V.

Originální název

UTILIZATION OF CONVOLUTIONAL NEURAL NETWORKS FOR SEGMENTATION OF MOUSE EMBRYOS CARTILAGINOUS TISSUE IN MICRO-CTDATA

Anglický název

UTILIZATION OF CONVOLUTIONAL NEURAL NETWORKS FOR SEGMENTATION OF MOUSE EMBRYOS CARTILAGINOUS TISSUE IN MICRO-CTDATA

Druh

Stať ve sborníku v databázi WoS či Scopus

Originální abstrakt

Automaticsegmentationofthebiologicalstructuresinmicro-CTdataisstillachallenge sincetheobjectofinterest(craniofacialcartilageinourcase)iscommonlynotcharacterizedbyunique voxelintensityorsharpborders.Inrecentyears,convolutionalneuralnetworks(CNNs)havebecome exceedinglypopularinmanyareasofcomputervision.Specifically,forbiomedicalimagesegmentationproblems,U-Netarchitectureiswidelyused.However,incaseofmicro-CTdata,thereis aquestionwhether3DCNNwouldnotbemorebeneficial.ThispaperintroducesCNNarchitecture basedonV-Netaswellasthemethodologyfordatapreprocessingandpostprocessing.Thebaseline architecturewasfurtheroptimizedusingadvancedtechniquessuchasAtrousSpatialPyramidPooling (ASPP)module,ScaledExponentialLinearUnit(SELU)activationfunction,multi-outputsupervisionandDenseblocks.Fornetworklearning,modernapproacheswereusedincludinglearningrate warmuporAdamWoptimizer.Eventhoughthe3DCNNdonotoutperformU-Netregardingthecraniofacialcartilagesegmentation,theoptimizationraisesthemedianofDicecoefficientfrom69.74% to80.01%.Therefore,utilizingtheseadvancedtechniquesishighlyencouragedastheycanbeeasily addedtoanyU-Net-likearchitectureandmayremarkablyimprovetheresults.

Anglický abstrakt

Automaticsegmentationofthebiologicalstructuresinmicro-CTdataisstillachallenge sincetheobjectofinterest(craniofacialcartilageinourcase)iscommonlynotcharacterizedbyunique voxelintensityorsharpborders.Inrecentyears,convolutionalneuralnetworks(CNNs)havebecome exceedinglypopularinmanyareasofcomputervision.Specifically,forbiomedicalimagesegmentationproblems,U-Netarchitectureiswidelyused.However,incaseofmicro-CTdata,thereis aquestionwhether3DCNNwouldnotbemorebeneficial.ThispaperintroducesCNNarchitecture basedonV-Netaswellasthemethodologyfordatapreprocessingandpostprocessing.Thebaseline architecturewasfurtheroptimizedusingadvancedtechniquessuchasAtrousSpatialPyramidPooling (ASPP)module,ScaledExponentialLinearUnit(SELU)activationfunction,multi-outputsupervisionandDenseblocks.Fornetworklearning,modernapproacheswereusedincludinglearningrate warmuporAdamWoptimizer.Eventhoughthe3DCNNdonotoutperformU-Netregardingthecraniofacialcartilagesegmentation,theoptimizationraisesthemedianofDicecoefficientfrom69.74% to80.01%.Therefore,utilizingtheseadvancedtechniquesishighlyencouragedastheycanbeeasily addedtoanyU-Net-likearchitectureandmayremarkablyimprovetheresults.

Klíčová slova

CNN,micro-CT,imagesegmentation,cartilaginoustissue,V-Net

Klíčová slova v angličtině

CNN,micro-CT,imagesegmentation,cartilaginoustissue,V-Net

Autoři

POLÁKOVÁ, V.

Rok RIV

2024

Vydáno

27.04.2021

Nakladatel

Brno University of Technology, Faculty of Electrical Engineering and Communication

Místo

Brno

ISBN

978-80-214-5943-4

Kniha

PROCEEDINGS II OF THE 27TH STUDENT EEICT 2021 selected papers

Edice

1

Strany od

106

Strany do

110

Strany počet

5

URL

https://www.eeict.cz/eeict_download/archiv/sborniky/EEICT_2021_sbornik_2_v3_DOI.pdf

BibTex

@inproceedings{BUT183914,
  author="Veronika {Poláková}",
  title="UTILIZATION OF CONVOLUTIONAL NEURAL NETWORKS FOR SEGMENTATION OF MOUSE EMBRYOS CARTILAGINOUS TISSUE IN MICRO-CTDATA",
  booktitle="PROCEEDINGS II OF THE 27TH STUDENT EEICT 2021 selected papers",
  year="2021",
  series="1",
  pages="106--110",
  publisher="Brno University of Technology, Faculty of Electrical Engineering and Communication",
  address="Brno",
  doi="10.13164/eeict.2021.106",
  isbn="978-80-214-5943-4",
  url="https://www.eeict.cz/eeict_download/archiv/sborniky/EEICT_2021_sbornik_2_v3_DOI.pdf"
}