Abstract
Although machine learning (ML) has shown promise across disciplines, out-of-sample generalizability is concerning. This is currently addressed by sharing multi-site data, but such centralization is challenging/infeasible to scale due to various limitations. Federated ML (FL) provides an alternative paradigm for accurate and generalizable ML, by only sharing numerical model updates. Here we present the largest FL study to-date, involving data from 71 sites across 6 continents, to generate an automatic tumor boundary detector for the rare disease of glioblastoma, reporting the largest such dataset in the literature (n = 6, 314). We demonstrate a 33% delineation improvement for the surgically targetable tumor, and 23% for the complete tumor extent, over a publicly trained model. We anticipate our study to: 1) enable more healthcare studies informed by large diverse data, ensuring meaningful results for rare diseases and underrepresented populations, 2) facilitate further analyses for glioblastoma by releasing our consensus model, and 3) demonstrate the FL effectiveness at such scale and task-complexity as a paradigm shift for multi-site collaborations, alleviating the need for data-sharing.
| Original language | English |
|---|---|
| Article number | 7346 |
| Journal | Nature Communications |
| Volume | 13 |
| Issue number | 1 |
| DOIs | |
| Publication status | Published - 5 Dec 2022 |
Fingerprint
Dive into the research topics of 'Federated learning enables big data for rare cancer boundary detection'. Together they form a unique fingerprint.Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver
}
In: Nature Communications, Vol. 13, No. 1, 7346, 05.12.2022.
Research output: Contribution to journal › Article › Research › peer-review
TY - JOUR
T1 - Federated learning enables big data for rare cancer boundary detection
AU - Pati, Sarthak
AU - Baid, Ujjwal
AU - Edwards, Brandon
AU - Sheller, Micah
AU - Wang, Shih Han
AU - Reina, G. Anthony
AU - Foley, Patrick
AU - Gruzdev, Alexey
AU - Karkada, Deepthi
AU - Davatzikos, Christos
AU - Sako, Chiharu
AU - Ghodasara, Satyam
AU - Bilello, Michel
AU - Mohan, Suyash
AU - Vollmuth, Philipp
AU - Brugnara, Gianluca
AU - Preetha, Chandrakanth J.
AU - Sahm, Felix
AU - Maier-Hein, Klaus
AU - Zenk, Maximilian
AU - Bendszus, Martin
AU - Wick, Wolfgang
AU - Calabrese, Evan
AU - Rudie, Jeffrey
AU - Villanueva-Meyer, Javier
AU - Cha, Soonmee
AU - Ingalhalikar, Madhura
AU - Jadhav, Manali
AU - Pandey, Umang
AU - Saini, Jitender
AU - Garrett, John
AU - Larson, Matthew
AU - Jeraj, Robert
AU - Currie, Stuart
AU - Frood, Russell
AU - Fatania, Kavi
AU - Huang, Raymond Y.
AU - Chang, Ken
AU - Quintero, Carmen Balaña
AU - Capellades, Jaume
AU - Puig, Josep
AU - Trenkler, Johannes
AU - Pichler, Josef
AU - Necker, Georg
AU - Haunschmidt, Andreas
AU - Meckel, Stephan
AU - Shukla, Gaurav
AU - Liem, Spencer
AU - Alexander, Gregory S.
AU - Lombardo, Joseph
AU - Palmer, Joshua D.
AU - Flanders, Adam E.
AU - Dicker, Adam P.
AU - Sair, Haris I.
AU - Jones, Craig K.
AU - Venkataraman, Archana
AU - Jiang, Meirui
AU - So, Tiffany Y.
AU - Chen, Cheng
AU - Heng, Pheng Ann
AU - Dou, Qi
AU - Kozubek, Michal
AU - Lux, Filip
AU - Michálek, Jan
AU - Matula, Petr
AU - Keřkovský, Miloš
AU - Kopřivová, Tereza
AU - Dostál, Marek
AU - Vybíhal, Václav
AU - Vogelbaum, Michael A.
AU - Mitchell, J. Ross
AU - Farinhas, Joaquim
AU - Maldjian, Joseph A.
AU - Yogananda, Chandan Ganesh Bangalore
AU - Pinho, Marco C.
AU - Reddy, Divya
AU - Holcomb, James
AU - Wagner, Benjamin C.
AU - Ellingson, Benjamin M.
AU - Cloughesy, Timothy F.
AU - Raymond, Catalina
AU - Oughourlian, Talia
AU - Hagiwara, Akifumi
AU - Wang, Chencai
AU - To, Minh Son
AU - Bhardwaj, Sargam
AU - Chong, Chee
AU - Agzarian, Marc
AU - Falcão, Alexandre Xavier
AU - Martins, Samuel B.
AU - Teixeira, Bernardo C.A.
AU - Sprenger, Flávia
AU - Menotti, David
AU - Lucio, Diego R.
AU - LaMontagne, Pamela
AU - Marcus, Daniel
AU - Wiestler, Benedikt
AU - Kofler, Florian
AU - Ezhov, Ivan
AU - Metz, Marie
AU - Jain, Rajan
AU - Lee, Matthew
AU - Lui, Yvonne W.
AU - McKinley, Richard
AU - Slotboom, Johannes
AU - Radojewski, Piotr
AU - Meier, Raphael
AU - Wiest, Roland
AU - Murcia, Derrick
AU - Fu, Eric
AU - Haas, Rourke
AU - Thompson, John
AU - Ormond, David Ryan
AU - Badve, Chaitra
AU - Sloan, Andrew E.
AU - Vadmal, Vachan
AU - Waite, Kristin
AU - Colen, Rivka R.
AU - Pei, Linmin
AU - Ak, Murat
AU - Srinivasan, Ashok
AU - Bapuraj, J. Rajiv
AU - Rao, Arvind
AU - Wang, Nicholas
AU - Yoshiaki, Ota
AU - Moritani, Toshio
AU - Turk, Sevcan
AU - Lee, Joonsang
AU - Prabhudesai, Snehal
AU - Morón, Fanny
AU - Mandel, Jacob
AU - Kamnitsas, Konstantinos
AU - Glocker, Ben
AU - Dixon, Luke V.M.
AU - Williams, Matthew
AU - Zampakis, Peter
AU - Panagiotopoulos, Vasileios
AU - Tsiganos, Panagiotis
AU - Alexiou, Sotiris
AU - Haliassos, Ilias
AU - Zacharaki, Evangelia I.
AU - Moustakas, Konstantinos
AU - Kalogeropoulou, Christina
AU - Kardamakis, Dimitrios M.
AU - Choi, Yoon Seong
AU - Lee, Seung Koo
AU - Chang, Jong Hee
AU - Ahn, Sung Soo
AU - Luo, Bing
AU - Poisson, Laila
AU - Wen, Ning
AU - Tiwari, Pallavi
AU - Verma, Ruchika
AU - Bareja, Rohan
AU - Yadav, Ipsa
AU - Chen, Jonathan
AU - Kumar, Neeraj
AU - Smits, Marion
AU - van der Voort, Sebastian R.
AU - Alafandi, Ahmed
AU - Incekara, Fatih
AU - Wijnenga, Maarten M.J.
AU - Kapsas, Georgios
AU - Gahrmann, Renske
AU - Schouten, Joost W.
AU - Dubbink, Hendrikus J.
AU - Vincent, Arnaud J.P.E.
AU - van den Bent, Martin J.
AU - French, Pim J.
AU - Klein, Stefan
AU - Yuan, Yading
AU - Sharma, Sonam
AU - Tseng, Tzu Chi
AU - Adabi, Saba
AU - Niclou, Simone P.
AU - Keunen, Olivier
AU - Hau, Ann Christin
AU - Vallières, Martin
AU - Fortin, David
AU - Lepage, Martin
AU - Landman, Bennett
AU - Ramadass, Karthik
AU - Xu, Kaiwen
AU - Chotai, Silky
AU - Chambless, Lola B.
AU - Mistry, Akshitkumar
AU - Thompson, Reid C.
AU - Gusev, Yuriy
AU - Bhuvaneshwar, Krithika
AU - Sayah, Anousheh
AU - Bencheqroun, Camelia
AU - Belouali, Anas
AU - Madhavan, Subha
AU - Booth, Thomas C.
AU - Chelliah, Alysha
AU - Modat, Marc
AU - Shuaib, Haris
AU - Dragos, Carmen
AU - Abayazeed, Aly
AU - Kolodziej, Kenneth
AU - Hill, Michael
AU - Abbassy, Ahmed
AU - Gamal, Shady
AU - Mekhaimar, Mahmoud
AU - Qayati, Mohamed
AU - Reyes, Mauricio
AU - Park, Ji Eun
AU - Yun, Jihye
AU - Kim, Ho Sung
AU - Mahajan, Abhishek
AU - Muzi, Mark
AU - Benson, Sean
AU - Beets-Tan, Regina G.H.
AU - Teuwen, Jonas
AU - Herrera-Trujillo, Alejandro
AU - Trujillo, Maria
AU - Escobar, William
AU - Abello, Ana
AU - Bernal, Jose
AU - Gómez, Jhon
AU - Choi, Joseph
AU - Baek, Stephen
AU - Kim, Yusung
AU - Ismael, Heba
AU - Allen, Bryan
AU - Buatti, John M.
AU - Kotrotsou, Aikaterini
AU - Li, Hongwei
AU - Weiss, Tobias
AU - Weller, Michael
AU - Bink, Andrea
AU - Pouymayou, Bertrand
AU - Shaykh, Hassan F.
AU - Saltz, Joel
AU - Prasanna, Prateek
AU - Shrestha, Sampurna
AU - Mani, Kartik M.
AU - Payne, David
AU - Kurc, Tahsin
AU - Pelaez, Enrique
AU - Franco-Maldonado, Heydy
AU - Loayza, Francis
AU - Quevedo, Sebastian
AU - Guevara, Pamela
AU - Torche, Esteban
AU - Mendoza, Cristobal
AU - Vera, Franco
AU - Ríos, Elvis
AU - López, Eduardo
AU - Velastin, Sergio A.
AU - Ogbole, Godwin
AU - Soneye, Mayowa
AU - Oyekunle, Dotun
AU - Odafe-Oyibotha, Olubunmi
AU - Osobu, Babatunde
AU - Shu’aibu, Mustapha
AU - Dorcas, Adeleye
AU - Dako, Farouk
AU - Simpson, Amber L.
AU - Hamghalam, Mohammad
AU - Peoples, Jacob J.
AU - Hu, Ricky
AU - Tran, Anh
AU - Cutler, Danielle
AU - Moraes, Fabio Y.
AU - Boss, Michael A.
AU - Gimpel, James
AU - Veettil, Deepak Kattil
AU - Schmidt, Kendall
AU - Bialecki, Brian
AU - Marella, Sailaja
AU - Price, Cynthia
AU - Cimino, Lisa
AU - Apgar, Charles
AU - Shah, Prashant
AU - Menze, Bjoern
AU - Barnholtz-Sloan, Jill S.
AU - Martin, Jason
AU - Bakas, Spyridon
N1 - Funding Information: Research and main methodological developments reported in this publication were partly supported by the National Institutes of Health (NIH) under award numbers NIH/NCI:U01CA242871 (S. Bakas), NIH/NINDS:R01NS042645 (C. Davatzikos), NIH/NCI:U24CA189523 (C. Davatzikos), NIH/NCI:U24CA215109 (J. Saltz), NIH/NCI:U01CA248226 (P. Tiwari), NIH/NCI:P30CA51008 (Y. Gusev), NIH:R50CA211270 (M. Muzi), NIH/NCATS:UL1TR001433 (Y. Yuan), NIH/NIBIB:R21EB030209 (Y. Yuan), NIH/NCI:R37CA214955 (A. Rao), and NIH:R01CA233888 (A.L. Simpson). The authors would also like to acknowledge the following NIH funded awards for the multi-site clinical trial (NCT00884741, RTOG0825/ACRIN6686): U10CA21661, U10CA37422, U10CA180820, U10CA180794, U01CA176110, R01CA082500, CA079778, CA080098, CA180794, CA180820, CA180822, CA180868. Research reported in this publication was also partly supported by the National Science Foundation, under award numbers 2040532 (S. Baek), and 2040462 (B. Landman). Research reported in this publication was also supported by i) a research grant from Varian Medical Systems (Palo Alto, CA, USA) (Y.Yuan), (ii) the Ministry of Health of the Czech Republic (Grant Nr. NU21-08-00359) (M.Kerkovský and M.Kozubek), (iii) Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) Project-ID 404521405, SFB 1389, Work Package C02, and Priority Program 2177 “Radiomics: Next Generation of Biomedical Imaging” (KI 2410/1-1 ∣ MA 6340/18-1) (P. Vollmuth), (iv) DFG Project-ID B12, SFB 824 (B. Wiestler), (v) the Helmholtz Association (funding number ZT-I-OO1 4) (K. Maier-Hein), vi) the Dutch Cancer Society (KWF project number EMCR 2015-7859) (S.R. van der Voort), (vii) the Chilean National Agency for Research and Development (ANID-Basal FB0008 (AC3E) and FB210017 (CENIA)) (P. Guevara), viii) the Canada CIFAR AI Chairs Program (M. Vallières), (ix) Leeds Hospital Charity (Ref: 9RO1/1403) (S. Currie), (x) the Cancer Research UK funding for the Leeds Radiotherapy Research Centre of Excellence (RadNet) and the grant number C19942/A28832 (S. Currie), (xi) Medical Research Council (MRC) Doctoral Training Program in Precision Medicine (Award Reference No. 2096671) (J. Bernal), (xii) The European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation program (Grant Agreement No. 757173) (B.Glocker), (xiii) The UKRI London Medical Imaging & Artificial Intelligence Centre for Value-Based Healthcare (K. Kamnitsas), (xiv) Wellcome/Engineering and Physical Sciences Research Council (EPSRC) Center for Medical Engineering (WT 203148/Z/16/Z) (T.C. Booth), (xv) American Cancer Society Research Scholar Grant RSG-16-005-01 (A. Rao), (xvi) the Department of Defense (DOD) Peer Reviewed Cancer Research Program (PRCRP) W81XWH-18-1-0404, Dana Foundation David Mahoney Neuroimaging Program, the V Foundation Translational Research Award, Johnson & Johnson WiSTEM2D Award (P. Tiwari), (xvii) RSNA Research & Education Foundation under grant number RR2011 (E.Calabrese), (xviii) the National Research Fund of Luxembourg (FNR) (grant number: C20/BM/14646004/GLASS-LUX/Niclou) (S.P.Niclou), xix) EU Marie Curie FP7-PEOPLE-2012-ITN project TRANSACT (PITN-GA-2012-316679) and the Swiss National Science Foundation (project number 140958) (J. Slotboom), and (xx) CNPq 303808/2018-7 and FAPESP 2014/12236-1 (A. Xavier Falcão). The content of this publication is solely the responsibility of the authors and does not represent the official views of the NIH, the NSF, the RSNA R&E Foundation, or any of the additional funding bodies. Publisher Copyright: © 2022, The Author(s).
PY - 2022/12/5
Y1 - 2022/12/5
N2 - Although machine learning (ML) has shown promise across disciplines, out-of-sample generalizability is concerning. This is currently addressed by sharing multi-site data, but such centralization is challenging/infeasible to scale due to various limitations. Federated ML (FL) provides an alternative paradigm for accurate and generalizable ML, by only sharing numerical model updates. Here we present the largest FL study to-date, involving data from 71 sites across 6 continents, to generate an automatic tumor boundary detector for the rare disease of glioblastoma, reporting the largest such dataset in the literature (n = 6, 314). We demonstrate a 33% delineation improvement for the surgically targetable tumor, and 23% for the complete tumor extent, over a publicly trained model. We anticipate our study to: 1) enable more healthcare studies informed by large diverse data, ensuring meaningful results for rare diseases and underrepresented populations, 2) facilitate further analyses for glioblastoma by releasing our consensus model, and 3) demonstrate the FL effectiveness at such scale and task-complexity as a paradigm shift for multi-site collaborations, alleviating the need for data-sharing.
AB - Although machine learning (ML) has shown promise across disciplines, out-of-sample generalizability is concerning. This is currently addressed by sharing multi-site data, but such centralization is challenging/infeasible to scale due to various limitations. Federated ML (FL) provides an alternative paradigm for accurate and generalizable ML, by only sharing numerical model updates. Here we present the largest FL study to-date, involving data from 71 sites across 6 continents, to generate an automatic tumor boundary detector for the rare disease of glioblastoma, reporting the largest such dataset in the literature (n = 6, 314). We demonstrate a 33% delineation improvement for the surgically targetable tumor, and 23% for the complete tumor extent, over a publicly trained model. We anticipate our study to: 1) enable more healthcare studies informed by large diverse data, ensuring meaningful results for rare diseases and underrepresented populations, 2) facilitate further analyses for glioblastoma by releasing our consensus model, and 3) demonstrate the FL effectiveness at such scale and task-complexity as a paradigm shift for multi-site collaborations, alleviating the need for data-sharing.
UR - http://www.scopus.com/inward/record.url?scp=85143349702&partnerID=8YFLogxK
UR - https://pubmed.ncbi.nlm.nih.gov/36470898
U2 - 10.1038/s41467-022-33407-5
DO - 10.1038/s41467-022-33407-5
M3 - Article
C2 - 36470898
AN - SCOPUS:85143349702
SN - 2041-1723
VL - 13
JO - Nature Communications
JF - Nature Communications
IS - 1
M1 - 7346
ER -