๐Ÿ”ฌ Medical Image/๋…ผ๋ฌธ ๋ฆฌ๋ทฐ

10๋…„๊ฐ„์˜ MICCAI BraTS Challenge data, task ์ด์ •๋ฆฌ

๋ณต๋งŒ 2021. 10. 8. 00:38

 

Brain Tumor Segmentation (BraTS) Challenge๋Š” MICCAI(Medical Image Computing and Computer Assisted Interventions)์—์„œ ์ฃผ์ตœํ•˜๋Š” challenge๋กœ, 2012๋…„์— ์ฒ˜์Œ ์‹œ์ž‘ํ•˜์—ฌ ์˜ฌํ•ด๋กœ 10์ฃผ๋…„์„ ๋งž์•˜๋‹ค.

 

๋ฐ์ดํ„ฐ์˜ ์ข…๋ฅ˜๋‚˜ lesion์˜ ์ข…๋ฅ˜ ๋“ฑ์— ๋”ฐ๋ผ brain tumor segmentation method์˜ ์„ฑ๋Šฅ์ด ๋‹ฌ๋ผ ํ‰๊ฐ€์— ์–ด๋ ค์›€์ด ์žˆ์–ด, state-of-the-art method๋ฅผ ์ฐพ๊ณ  large public dataset์„ ๋งŒ๋“ค๊ธฐ ์œ„ํ•ด ์‹œ์ž‘๋˜์—ˆ๋‹ค๊ณ  ํ•œ๋‹ค.

 

๋งค๋…„ Brain tumor๊ณผ ๊ด€๋ จ๋œ ๋ฐ์ดํ„ฐ์…‹๊ณผ task๊ฐ€ ๊ณต๊ฐœ๋˜์–ด, ๋งŽ์€ ์—ฐ๊ตฌ์—์„œ๋„ ์ด์šฉ๋˜๊ณ  ์žˆ๋‹ค.

 

์˜ฌํ•ด๋Š” RNSA(Radiological Society of North America), ASNR(Americal Society of Neuroradiology), MICCAI๊ฐ€ ๊ณต๋™ ์ฃผ์ตœํ•œ๋‹ค.

 

์˜ฌํ•ด๋ฅผ ํฌํ•จํ•ด ์ด 10ํšŒ์˜ BraTS Challenge์˜ ๋ฐ์ดํ„ฐ์…‹๊ณผ task, ๊ฒฐ๊ณผ๋ฅผ ๊ฐ„๋‹จํžˆ ์ •๋ฆฌํ•ด ๋ณด๊ณ ์ž ํ•œ๋‹ค.

 


 

BraTS 2012

Home: http://www2.imm.dtu.dk/projects/BRATS2012/

Proceedings: https://www.cbica.upenn.edu/sbia/Spyridon.Bakas/MICCAI_BraTS/MICCAI_BraTS_2012_proceedings.pdf

 

Dataset & Task

Train data๋กœ 30๋ช…์˜ glioma(๋‡Œ๊ต์ข…) ํ™˜์ž์˜ T1, T2, FLAIR, post-Gadolinium T1 4๊ฐ€์ง€ contrast์˜ MRI๊ฐ€ ์ œ๊ณต๋˜์—ˆ์œผ๋ฉฐ, ์ถ”๊ฐ€๋กœ 50๋ช…์˜ synthetic data ์—ญ์‹œ ์ œ๊ณต๋˜์—ˆ๋‹ค.

 

Task๋Š” lesion segmentation์œผ๋กœ, ๊ฐ data์—๋Š” "active tumor"์™€ "edema"๊ฐ€ label๋กœ ์ฃผ์–ด์ ธ ์žˆ์–ด, ์ด๋“ค์„ segmentationํ•˜๋Š” ๊ฒƒ์ด ๋ชฉ์ ์ด๋‹ค.

 

Result

์ด 12ํŒ€์ด accept๋˜์—ˆ์œผ๋ฉฐ, ๋Œ€๋ถ€๋ถ„์ด Machine learning์„ ์‚ฌ์šฉํ–ˆ๋‹ค.

 

D. Zikic et al., Context-sensitive Classification Forests for Segmentation of Brain Tumor Tissues

 


 

BraTS 2013

Home: http://martinos.org/qtim/miccai2013/

Proceedings: https://www.cbica.upenn.edu/sbia/Spyridon.Bakas/MICCAI_BraTS/MICCAI_BraTS_2013_proceedings.pdf

 

Dataset & Task

BraTS 2012์™€ data์™€ task ๋ชจ๋‘ ๋™์ผํ•˜๋‚˜, ํ™˜์ž 20๋ช…์˜ data๊ฐ€ ์ถ”๊ฐ€๋กœ ๋”ํ•ด์กŒ๋‹ค.

์ด 50๋ช…์˜ glioma ํ™˜์ž์˜ T1, T2, FLAIR, post-Gadolinium T1 4๊ฐ€์ง€ contrast์˜ MRI๊ฐ€ ์ œ๊ณต๋˜์—ˆ์œผ๋ฉฐ,

50๋ช…์˜ synthetic data ์—ญ์‹œ ์ œ๊ณต๋˜์—ˆ๋‹ค.

 

Task ์—ญ์‹œ ์ด์ „๊ณผ ๋™์ผํ•œ lesion segmentation์ด๊ณ , label๋กœ๋Š” "edema", "enhancing core", "non-enhancing core", "necrotic core" ๋„ค ๊ฐ€์ง€๊ฐ€ ์ฃผ์–ด์กŒ๋‹ค.

 


 

BraTS 2014

Home: https://sites.google.com/site/miccaibrats2014/

Proceedings: https://www.cbica.upenn.edu/sbia/Spyridon.Bakas/MICCAI_BraTS/MICCAI_BraTS_2014_proceedings.pdf

 

Dataset & Task

์ด์ „๊นŒ์ง€์˜ challenge์™€ ๋‹ค๋ฅด๊ฒŒ, ์ด 3๊ฐœ์˜ task๊ฐ€ ์ฃผ์–ด์กŒ๋‹ค.

 

Task 1: Lesion Segmentation

BraTS 2012, 2013์—์„œ ์ง„ํ–‰ํ•œ lesion segmentation์ด๋‹ค. ์ด์ „๊ณผ ๋‹ค๋ฅด๊ฒŒ ํ›จ์”ฌ ๋งŽ์•„์ง„ 300๋ช…์˜ T1, T1 contrast-enhanced, T2, T2 FLAIR MRI๊ฐ€ ์ฃผ์–ด์กŒ๋‹ค. BraTS 2012, 2013์˜ ๋ฐ์ดํ„ฐ์— NIH Cancer Imaging Archieve (TCIA)๋ฅผ ๋”ํ•ด ๋งŒ๋“ค์–ด์กŒ๋‹ค. Label์€ ์ด์ „๊ณผ ๋™์ผํ•˜๋‹ค.

 

Task 2: Longtitudinal Lesion Segmentation

3-10๊ฐœ์˜ time point๋ฅผ ๊ฐ€์ง€๊ณ  ์žˆ๋Š” 25๋ช…์˜ ํ™˜์ž data๊ฐ€ ์ฃผ์–ด์ง€๊ณ , ์ด๋ฅผ ์ด์šฉํ•ด lesion segmentation์„ ํ•˜๋Š” task์ด๋‹ค.

 

Task 3: Diagnostic Image Classification

์ฒ˜์Œ์œผ๋กœ classification task๊ฐ€ ๋“ฑ์žฅํ•œ๋‹ค. 

 

Result

๊ฐ„๋‹จํ•œ CNN์„ ์‚ฌ์šฉํ•œ ํŒ€๋„ ์žˆ๊ณ , ๋Œ€๋ถ€๋ถ„ machine learning์„ ์‚ฌ์šฉํ–ˆ๋‹ค.

Axel Davy et al., Brain Tumor Segmentation with Deep Neural Networks

 


 

BraTS 2015

Home: https://sites.google.com/site/braintumorsegmentation/home/brats2015

Proceedings: https://www.cbica.upenn.edu/sbia/Spyridon.Bakas/MICCAI_BraTS/MICCAI_BraTS_2015_proceedings.pdf

 

Dataset & Task

Task๋Š” ์ด์ „๊ณผ ๋™์ผํ•˜๊ฒŒ lesion segmentation์ด๊ณ , ๋ฐ์ดํ„ฐ๋Š” BraTS 2014์™€ ๋™์ผํ•˜๋ฉฐ test data๋งŒ ์ƒˆ๋กญ๊ฒŒ ์ค€๋น„๋˜์—ˆ๋‹ค๊ณ  ํ•œ๋‹ค.

 

Result

์ด์ „๊ณผ ๋งˆ์ฐฌ๊ฐ€์ง€๋กœ ๋Œ€๋ถ€๋ถ„ CNN ํ˜น์€ machine learning์„ ์ด์šฉํ–ˆ๋‹ค.

 


 

BraTS 2016

Home: https://sites.google.com/site/braintumorsegmentation/home/brats_2016

Proceedings: https://www.cbica.upenn.edu/sbia/Spyridon.Bakas/MICCAI_BraTS/MICCAI_BraTS_2016_proceedings.pdf

 

Dataset & Task

๋ฐ์ดํ„ฐ๋Š” BraTS 2014, 2015์™€ ๋™์ผํ•˜๋‹ค. Task๋Š” ๋‘ ๊ฐ€์ง€๊ฐ€ ์žˆ๋Š”๋ฐ ํ•˜๋‚˜๋Š” ๊ณ„์† ์ง„ํ–‰ํ•ด ์˜จ lesion segmentation์ด๊ณ , ๋‹ค๋ฅธ ํ•˜๋‚˜๋Š” "Quantifying longitudinal changes"์ด๋‹ค. ์ด๋Š” ๊ฐ™์€ ํ™˜์ž์— ๋Œ€ํ•ด ์˜ˆํ›„๋ฅผ ๊ด€์ธกํ•˜๊ธฐ ์œ„ํ•œ 2ํšŒ์˜ ์ถ”๊ฐ€ ์ดฌ์˜์„ ํ•˜๊ณ , ์ฐธ๊ฐ€์ž๋“ค์˜ segmentation ๊ฒฐ๊ณผ๊ฐ€ ์˜ˆํ›„๋ฅผ ์ธก์ •ํ•  ์ˆ˜ ์žˆ์„ ์ •๋„๋กœ ์œ ์˜๋ฏธํ•œ ์ฐจ์ด๋ฅผ ๋ณด์ด๋Š”์ง€๋ฅผ ํ‰๊ฐ€ํ•˜๋Š” ๊ฒƒ์ด๋‹ค.

 

Result

์ด์ „๊ณผ ๋งˆ์ฐฌ๊ฐ€์ง€๋กœ ๋Œ€๋ถ€๋ถ„ CNN ํ˜น์€ machine learning์„ ์ด์šฉํ–ˆ๋‹ค.

 

Petet D. Chang, Fully Convolutional Neural Networks with Hyperlocal Features for Brain Tumor Segmentation

 


 

BraTS 2017

Home: https://www.med.upenn.edu/cbica/aibil/srathore.html

Proceedings: https://www.cbica.upenn.edu/sbia/Spyridon.Bakas/MICCAI_BraTS/MICCAI_BraTS_2017_proceedings_shortPapers.pdf

 

Dataset & Task

BraTS 2014~2016์—์„œ ์‚ฌ์šฉ๋œ TCIA dataset์„ ์ œ์™ธํ–ˆ๋‹ค. ์ด์œ ๋Š” 1) TCIA dataset์ด pre-operative์™€ post-operative data๋ฅผ ํฌํ•จํ•˜๊ณ  ์žˆ๊ณ , 2) Label์ด ์ง์ ‘ annotated๋œ ๊ฒƒ์ด ์•„๋‹Œ, BraTS 2012, 2013์—์„œ ๋†’์€ ์„ฑ๋Šฅ์„ ๋ณด์ธ method๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ํ•˜์—ฌ ์ƒ์„ฑํ–ˆ๊ธฐ ๋•Œ๋ฌธ์ด๋‹ค. 

์ด๋Ÿฌํ•œ ์ด์œ ๋กœ TCIA ๋ฐ์ดํ„ฐ์…‹์—์„œ pre-operative data๋งŒ์„ ๊ณจ๋ผ๋‚ด์–ด, ๋‹ค์‹œ label์„ ์ง์ ‘ annotationํ•˜๋Š” ์ž‘์—…์„ ๊ฑฐ์ณ BraTS 2012, 2013์˜ ๋ฐ์ดํ„ฐ์™€ ํ•จ๊ป˜ ์‚ฌ์šฉํ•˜์—ฌ ์ด 285๊ฐœ์˜ data๊ฐ€ ๋˜์—ˆ๋‹ค.

 

Task๋Š” ๋‹ค์Œ ๋‘ ๊ฐ€์ง€์ด๋‹ค.

 

Task 1: Segmentation of gliomas in pre-operative scans

Enhancing Tumor (ET), Tumor Core (TC), Whole Tumor (WT)๋ฅผ segmentationํ•˜๋Š” task์ด๋‹ค.

 

Task 2: Prediction of patient overall survival (OS) from pre-operative scans

์ฃผ์–ด์ง„ data์™€ segmentation label๋“ค์„ ์ด์šฉํ•ด ์ƒ์กด๋ฅ ์„ ์˜ˆ์ธกํ•˜๋Š” task์ด๋‹ค. Patient์˜ ๋‚˜์ด ๋“ฑ์˜ ์ •๋ณด๋ฅผ ํ•จ๊ป˜ ํ™œ์šฉํ•  ์ˆ˜ ์žˆ๋‹ค.

classification๊ณผ regression ๋‘ ๊ฐ€์ง€์˜ ๊ด€์ ์—์„œ ์„ฑ๋Šฅ์„ ํ‰๊ฐ€ํ–ˆ๋‹ค.

Classification์€ 3๊ฐ€์ง€์˜ category (long-survivers, short-survivors, mid-survivors)๋กœ ๋‚˜๋‰˜๊ณ ,

regression์€ pairwise๋กœ MSE๋ฅผ ์ธก์ •ํ–ˆ๋‹ค.

 

Result

Segmentation์€ ๋Œ€๋ถ€๋ถ„ U-net ๊ณ„์—ด ๊ตฌ์กฐ๋ฅผ ์‚ฌ์šฉํ–ˆ๊ณ , OS prediction์˜ ๊ฒฝ์šฐ SVM, XGBoost ๋“ฑ machine learning์„ ๋Œ€๋ถ€๋ถ„ ์‚ฌ์šฉํ–ˆ๋‹ค.

 

Varghese Alex et al., Brain Tumor Segmentation from Multi Modal MR images using Fully Convolutional Neural Network

 


 

BraTS 2018

Home: http://braintumorsegmentation.org/

Proceedings: https://www.cbica.upenn.edu/sbia/Spyridon.Bakas/MICCAI_BraTS/MICCAI_BraTS_2018_proceedings_shortPapers.pdf

 

Dataset & Task

BraTS 2017๊ณผ dataset, task๊ฐ€ ๋™์ผํ•˜๋‹ค.

 

Result

์ด์ „๊ณผ ์œ ์‚ฌํ•˜๊ฒŒ segmentation์€ ๋Œ€๋ถ€๋ถ„ U-net ๊ณ„์—ด ๊ตฌ์กฐ๋ฅผ ์‚ฌ์šฉํ–ˆ๊ณ , OS prediction์˜ ๊ฒฝ์šฐ ๋‹ค์–‘ํ•œ machine learning์„ ์ด์šฉํ–ˆ๋‹ค.

 

Subhashis Banerjee et al., Multi-Planar Spatial-ConvNet for Segmentation and Survival Prediction in Brain Cancer

 


 

BraTS 2019

Home: http://braintumorsegmentation.org/

Proceedings: https://link.springer.com/book/10.1007/978-3-030-46640-4 (vol 1), https://link.springer.com/book/10.1007/978-3-030-46643-5 (vol 2)

 

Dataset & Task

BraTS 2017๊ณผ ๋™์ผํ•˜๊ฒŒ lesion segmentation, OS prediction์„ ์ง„ํ–‰ํ•˜๋ฉฐ, "Quantification of Uncertainty in Segmentation" task๊ฐ€ ์ถ”๊ฐ€๋˜์—ˆ๋‹ค.

์ด๋Š” segmentation์˜ uncertainty๋ฅผ ์ •๋Ÿ‰ํ™”ํ•˜๋Š” ๋ฐฉ๋ฒ•์„ ์ œ์•ˆํ•˜๋Š” ๊ฒƒ์œผ๋กœ, 0-100์˜ ๋ฒ”์œ„๋กœ ๊ฐ์ž์˜ segmentation ๊ฒฐ๊ณผ์˜ uncertainty๋ฅผ ์ธก์ •ํ•˜๊ฒŒ ๋œ๋‹ค.

 

Result

-

 


 

BraTS 2020

Home: https://www.med.upenn.edu/cbica/brats2020/

Proceedings: https://link.springer.com/book/10.1007/978-3-030-72084-1 (vol 1), https://link.springer.com/book/10.1007/978-3-030-72087-2 (vol 2)

 

Dataset & Task

BraTS 2019์™€ ๋™์ผํ•˜๊ฒŒ 3๊ฐ€์ง€ task๊ฐ€ ์ฃผ์–ด์กŒ๋‹ค.

 

Result

-

 


 

BraTS 2021 (In Progress)

Home: http://braintumorsegmentation.org/

 

Dataset & Task

BraTS 2020์—์„œ ์‚ฌ์šฉ๋œ 660๋ช…์˜ ๋ฐ์ดํ„ฐ๋ฅผ ํฌํ•จํ•˜์—ฌ, 2,000๋ช…์˜ ๋ฐ์ดํ„ฐ๋ฅผ ์ œ๊ณตํ•œ๋‹ค.

 

Task๋Š” ๋งค๋ฒˆ ์ง„ํ–‰๋˜๋˜ lesion segmentation์— ๋”ํ•˜์—ฌ classification task๊ฐ€ ์ถ”๊ฐ€๋˜์—ˆ๋‹ค.

Classification task๋Š” MGMT promoter methylation status๋ฅผ ์˜ˆ์ธกํ•˜๋Š” ๊ฒƒ์ด๋‹ค. 

MGMT promoter methylation๋Š” ์ข…์–‘์—์„œ ๋‚˜ํƒ€๋‚˜๋Š” ํŠน์ •ํ•œ ์œ ์ „์ž ์—ผ๊ธฐ์„œ์—ด ์ค‘ ํ•˜๋‚˜๋กœ, ์˜ˆํ›„ ์ง„๋‹จ๊ณผ ํ™”ํ•™์š”๋ฒ•์˜ ํ•„์š”์—ฌ๋ถ€๋ฅผ ๊ฒฐ์ •ํ•˜๋Š” ๋ฐ์— ์ค‘์š”ํ•œ ์—ญํ• ์„ ํ•œ๋‹ค.

๋‹จ, ์ด๋Ÿฌํ•œ ์œ ์ „์  ๋ถ„์„์„ ํ•˜๊ธฐ ์œ„ํ•ด์„œ๋Š” ์ˆ˜์ˆ ์ด ํ•„์š”ํ•˜๊ณ , ๊ฒฐ๊ณผ๋ฅผ ์–ป๋Š” ๋ฐ์—๋„ ๋งŽ์€ ์‹œ๊ฐ„์ด ์†Œ์š”๋˜๋Š”๋ฐ, ์˜์ƒ ์ •๋ณด๋งŒ์„ ์ด์šฉํ•˜์—ฌ ์ด๋ฅผ ํ™•์ธํ•  ์ˆ˜ ์žˆ๋‹ค๋ฉด ์ž„์ƒ์ ์œผ๋กœ ํฐ ์˜๋ฏธ๊ฐ€ ์žˆ์„ ๊ฒƒ์ด๋ผ๊ณ  ํ•œ๋‹ค.

MGMT promoter methylation classification์— ๋Œ€ํ•œ ๋”์šฑ ์ž์„ธํ•œ ์ •๋ณด๋Š” ์—ฌ๊ธฐ์—์„œ ํ™•์ธํ•  ์ˆ˜ ์žˆ๋‹ค.

 

Result

-

 

 

๋ฐ˜์‘ํ˜•