Automatic segmentation of the pulmonary lobes with a 3D u-net and optimized loss function

Bianca Lassen-Schmidt; Alessa Denise Hering; Stefan Krass; Hans  Meine

doi:10.48550/arXiv.2006.00083

Automatic segmentation of the pulmonary lobes with a 3D u-net and optimized loss function

Bianca Lassen-Schmidt, Alessa Denise Hering, Stefan Krass, Hans Meine

Institute of Mathematics and Image Computing

Fraunhofer MEVIS, Max-von-Laue-Str. 2, 28359 Bremen

Abstract

Fully-automatic lung lobe segmentation is challenging due to anatomical variations, pathologies, and incomplete fissures. We trained a 3D u-net for pulmonary lobe segmentation on 49 mainly publically available datasets and introduced a weighted Dice loss function to emphasize the lobar boundaries. To validate the performance of the proposed method we compared the results to two other methods. The new loss function improved the mean distance to 1.46 mm (compared to 2.08 mm for simple loss function without weighting).

Original language	English
DOIs	https://doi.org/10.48550/arXiv.2006.00083
Publication status	Published - 29.05.2020
Event	Medical Imaging with Deep Learning 2020 - Montréal, Montréal, Canada Duration: 06.07.2020 → 09.07.2020 https://2020.midl.io/

Conference

Conference	Medical Imaging with Deep Learning 2020
Abbreviated title	MIDL 2020
Country/Territory	Canada
City	Montréal
Period	06.07.20 → 09.07.20
Internet address	https://2020.midl.io/

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

Access to Document

10.48550/arXiv.2006.00083

Cite this

@conference{1dc1c9fca8df46efb7c71b7437f96b3c,

title = "Automatic segmentation of the pulmonary lobes with a 3D u-net and optimized loss function",

abstract = "Fully-automatic lung lobe segmentation is challenging due to anatomical variations, pathologies, and incomplete fissures. We trained a 3D u-net for pulmonary lobe segmentation on 49 mainly publically available datasets and introduced a weighted Dice loss function to emphasize the lobar boundaries. To validate the performance of the proposed method we compared the results to two other methods. The new loss function improved the mean distance to 1.46 mm (compared to 2.08 mm for simple loss function without weighting). ",

author = "Bianca Lassen-Schmidt and Hering, {Alessa Denise} and Stefan Krass and Hans Meine",

year = "2020",

month = may,

day = "29",

doi = "10.48550/arXiv.2006.00083",

language = "English",

note = "Medical Imaging with Deep Learning 2020 ; Conference date: 06-07-2020 Through 09-07-2020",

url = "https://2020.midl.io/",

}

TY - CONF

T1 - Automatic segmentation of the pulmonary lobes with a 3D u-net and optimized loss function

AU - Lassen-Schmidt, Bianca

AU - Hering, Alessa Denise

AU - Krass, Stefan

AU - Meine, Hans

PY - 2020/5/29

Y1 - 2020/5/29

N2 - Fully-automatic lung lobe segmentation is challenging due to anatomical variations, pathologies, and incomplete fissures. We trained a 3D u-net for pulmonary lobe segmentation on 49 mainly publically available datasets and introduced a weighted Dice loss function to emphasize the lobar boundaries. To validate the performance of the proposed method we compared the results to two other methods. The new loss function improved the mean distance to 1.46 mm (compared to 2.08 mm for simple loss function without weighting).

AB - Fully-automatic lung lobe segmentation is challenging due to anatomical variations, pathologies, and incomplete fissures. We trained a 3D u-net for pulmonary lobe segmentation on 49 mainly publically available datasets and introduced a weighted Dice loss function to emphasize the lobar boundaries. To validate the performance of the proposed method we compared the results to two other methods. The new loss function improved the mean distance to 1.46 mm (compared to 2.08 mm for simple loss function without weighting).

U2 - 10.48550/arXiv.2006.00083

DO - 10.48550/arXiv.2006.00083

M3 - Conference Papers

T2 - Medical Imaging with Deep Learning 2020

Y2 - 6 July 2020 through 9 July 2020

ER -