PaperlayerView exact revision on arXiv

Source-linked AI summary

Federated Learning Enables Big Data for Rare Cancer Boundary Detection

Sarthak Pati, Ujjwal Baid, Brandon Edwards, Micah Sheller, Shih-Han Wang, G Anthony Reina, Patrick Foley, Alexey Gruzdev, Deepthi Karkada, Christos Davatzikos, Chiharu Sako, Satyam Ghodasara, Michel Bilello, Suyash Mohan, Philipp Vollmuth, Gianluca Brugnara, Chandrakanth J Preetha, Felix Sahm, Klaus Maier-Hein, Maximilian Zenk, Martin Bendszus, Wolfgang Wick, Evan Calabrese, Jeffrey Rudie, Javier Villanueva-Meyer, Soonmee Cha, Madhura Ingalhalikar, Manali Jadhav, Umang Pandey, Jitender Saini, John Garrett, Matthew Larson, Robert Jeraj, Stuart Currie, Russell Frood, Kavi Fatania, Raymond Y Huang, Ken Chang, Carmen Balana, Jaume Capellades, Josep Puig, Johannes Trenkler, Josef Pichler, Georg Necker, Andreas Haunschmidt, Stephan Meckel, Gaurav Shukla, Spencer Liem, Gregory S Alexander, Joseph Lombardo, Joshua D Palmer, Adam E Flanders, Adam P Dicker, Haris I Sair, Craig K Jones, Archana Venkataraman, Meirui Jiang, Tiffany Y So, Cheng Chen, Pheng Ann Heng, Qi Dou, Michal Kozubek, Filip Lux, Jan Michálek, Petr Matula, Miloš Keřkovský, Tereza Kopřivová, Marek Dostál, Václav Vybíhal, Michael A Vogelbaum, J Ross Mitchell, Joaquim Farinhas, Joseph A Maldjian, Chandan Ganesh Bangalore Yogananda, Marco C Pinho, Divya Reddy, James Holcomb, Benjamin C Wagner, Benjamin M Ellingson, Timothy F Cloughesy, Catalina Raymond, Talia Oughourlian, Akifumi Hagiwara, Chencai Wang, Minh-Son To, Sargam Bhardwaj, Chee Chong, Marc Agzarian, Alexandre Xavier Falcão, Samuel B Martins, Bernardo C A Teixeira, Flávia Sprenger, David Menotti, Diego R Lucio, Pamela LaMontagne, Daniel Marcus, Benedikt Wiestler, Florian Kofler, Ivan Ezhov, Marie Metz, Rajan Jain, Matthew Lee, Yvonne W Lui, Richard McKinley, Johannes Slotboom, Piotr Radojewski, Raphael Meier, Roland Wiest, Derrick Murcia, Eric Fu, Rourke Haas, John Thompson, David Ryan Ormond, Chaitra Badve, Andrew E Sloan, Vachan Vadmal, Kristin Waite, Rivka R Colen, Linmin Pei, Murat Ak, Ashok Srinivasan, J Rajiv Bapuraj, Arvind Rao, Nicholas Wang, Ota Yoshiaki, Toshio Moritani, Sevcan Turk, Joonsang Lee, Snehal Prabhudesai, Fanny Morón, Jacob Mandel, Konstantinos Kamnitsas, Ben Glocker, Luke V M Dixon, Matthew Williams, Peter Zampakis, Vasileios Panagiotopoulos, Panagiotis Tsiganos, Sotiris Alexiou, Ilias Haliassos, Evangelia I Zacharaki, Konstantinos Moustakas, Christina Kalogeropoulou, Dimitrios M Kardamakis, Yoon Seong Choi, Seung-Koo Lee, Jong Hee Chang, Sung Soo Ahn, Bing Luo, Laila Poisson, Ning Wen, Pallavi Tiwari, Ruchika Verma, Rohan Bareja, Ipsa Yadav, Jonathan Chen, Neeraj Kumar, Marion Smits, Sebastian R van der Voort, Ahmed Alafandi, Fatih Incekara, Maarten MJ Wijnenga, Georgios Kapsas, Renske Gahrmann, Joost W Schouten, Hendrikus J Dubbink, Arnaud JPE Vincent, Martin J van den Bent, Pim J French, Stefan Klein, Yading Yuan, Sonam Sharma, Tzu-Chi Tseng, Saba Adabi, Simone P Niclou, Olivier Keunen, Ann-Christin Hau, Martin Vallières, David Fortin, Martin Lepage, Bennett Landman, Karthik Ramadass, Kaiwen Xu, Silky Chotai, Lola B Chambless, Akshitkumar Mistry, Reid C Thompson, Yuriy Gusev, Krithika Bhuvaneshwar, Anousheh Sayah, Camelia Bencheqroun, Anas Belouali, Subha Madhavan, Thomas C Booth, Alysha Chelliah, Marc Modat, Haris Shuaib, Carmen Dragos, Aly Abayazeed, Kenneth Kolodziej, Michael Hill, Ahmed Abbassy, Shady Gamal, Mahmoud Mekhaimar, Mohamed Qayati, Mauricio Reyes, Ji Eun Park, Jihye Yun, Ho Sung Kim, Abhishek Mahajan, Mark Muzi, Sean Benson, Regina G H Beets-Tan, Jonas Teuwen, Alejandro Herrera-Trujillo, Maria Trujillo, William Escobar, Ana Abello, Jose Bernal, Jhon Gómez, Joseph Choi, Stephen Baek, Yusung Kim, Heba Ismael, Bryan Allen, John M Buatti, Aikaterini Kotrotsou, Hongwei Li, Tobias Weiss, Michael Weller, Andrea Bink, Bertrand Pouymayou, Hassan F Shaykh, Joel Saltz, Prateek Prasanna, Sampurna Shrestha, Kartik M Mani, David Payne, Tahsin Kurc, Enrique Pelaez, Heydy Franco-Maldonado, Francis Loayza, Sebastian Quevedo, Pamela Guevara, Esteban Torche, Cristobal Mendoza, Franco Vera, Elvis Ríos, Eduardo López, Sergio A Velastin, Godwin Ogbole, Dotun Oyekunle, Olubunmi Odafe-Oyibotha, Babatunde Osobu, Mustapha Shu'aibu, Adeleye Dorcas, Mayowa Soneye, Farouk Dako, Amber L Simpson, Mohammad Hamghalam, Jacob J Peoples, Ricky Hu, Anh Tran, Danielle Cutler, Fabio Y Moraes, Michael A Boss, James Gimpel, Deepak Kattil Veettil, Kendall Schmidt, Brian Bialecki, Sailaja Marella, Cynthia Price, Lisa Cimino, Charles Apgar, Prashant Shah, Bjoern Menze, Jill S Barnholtz-Sloan, Jason Martin, Spyridon Bakas

arXiv:2204.10836v2cs.LGeess.IV

TL;DR

The study addresses the challenge of developing generalizable glioblastoma boundary-detection models from large, diverse, multisite data. It uses federated learning across 71 sites, improving performance over a public initial model on local validation and complete out-of-sample data.

  • Problem

    Generalizable glioblastoma boundary detection requires large and diverse data from multiple sites, whose use is constrained by centralized-data limitations.

  • Method

    Federated learning combined data from 71 sites and 6,314 cases to develop a glioblastoma sub-compartment boundary-detection model without centralizing collaborators’ data.

  • Results

    The consensus model improved performance over the public initial model by 27% for ET, 33% for TC, and 16% for WT on local validation, and by 15%, 27%, and 16%, respectively, on complete out-of-sample data.

  • Takeaways & Limitations

    Federated learning enabled an accurate and generalizable model relevant to neurosurgical and radiotherapy planning and provides a blueprint for future clinically deployable FL studies.

  • Takeaways & Limitations

    The model was designed as a single 3D-ResUNet to reduce computational burden in low-resource clinical environments rather than using an ensemble.

Abstract

from arXiv · show

Although machine learning (ML) has shown promise in numerous domains, there are concerns about generalizability to out-of-sample data. This is currently addressed by centrally sharing ample, and importantly diverse, data from multiple sites. However, such centralization is challenging to scale (or even not feasible) due to various limitations. Federated ML (FL) provides an alternative to train accurate and generalizable ML models, by only sharing numerical model updates. Here we present findings from the largest FL study to-date, involving data from 71 healthcare institutions across 6 continents, to generate an automatic tumor boundary detector for the rare disease of glioblastoma, utilizing the largest dataset of such patients ever used in the literature (25,256 MRI scans from 6,314 patients). We demonstrate a 33% improvement over a publicly trained model to delineate the surgically targetable tumor, and 23% improvement over the tumor's entire extent. We anticipate our study to: 1) enable more studies in healthcare informed by large and diverse data, ensuring meaningful results for rare diseases and underrepresented populations, 2) facilitate further quantitative analyses for glioblastoma via performance optimization of our consensus model for eventual public release, and 3) demonstrate the effectiveness of FL at such scale and task complexity as a paradigm shift for multi-site collaborations, alleviating the need for data sharing.

Methods

The study federated glioblastoma MRI data across geographically distinct sites while keeping training data local and sharing model updates through a central aggregation server. It used harmonized preprocessing and a single 3D-ResUNet designed for deployment in resource-constrained clinical environments.

  • Data: Each case required native T1, gadolinium-enhanced T1, T2, and T2-FLAIR MRI sequences, with missing-sequence cases excluded.Both 2D axial and 3D acquisitions were included, with a preference for 3D when available.
  • Data: Six sites were held out as out-of-sample collaborators because they were not part of the training stage.These sites were allocated for model generalizability validation.
  • Data: 71 sites contributed to a final consensus model trained on 6,314 glioblastoma cases.The federation included 55 initial collaborating sites, with additional participation producing the final 71-site dataset.
  • Preprocessing: Sites applied BraTS-based harmonized preprocessing to account for acquisition differences such as 2D versus 3D scans.The provided platform supported preprocessing before local training.
  • Model: The tumor-boundary detector used a 3D-ResUNet with residual connections, 30 base filters, and Adam optimization at lr = 5 × 10^-5.Training used generalized DSC loss applied independently to the absolute complement of each tumor sub-compartment.
  • Federated training: In each federated round, sites trained the shared architecture locally for one epoch and sent model updates to a central aggregation server.The federation began from a public initial model trained on 231 cases from 16 sites, rather than random initialization.
  • Clinical deployment: A single 3D-ResUNet was selected instead of an ensemble to reduce computational burden in low-resource clinical settings.The planned optimized release used graph-level and 8-bit quantization optimizations for lower latency and memory requirements.
  • Code availability: The FeTS platform integrated preprocessing, annotation refinement, segmentation, label fusion, data loading, augmentation, and federated-learning components from open-source tools.The study released the design code and incorporated tools including CaPTk, BrainMaGe, DeepMedic, nnU-Net, GaNDLF, PyTorch, and TorchIO.

Results

The final consensus model outperformed the public initial model on both collaborators’ local validation data and complete out-of-sample data. These gains followed federated training on the larger, diverse collaborator dataset.

  • Baseline: The public initial model had average DSC values of 0.63 for ET, 0.62 for TC, and 0.75 for WT, with a collective average DSC of 0.66.These values were measured across all cases and sites at federation initialization.
  • Local validation: 33% improvement was achieved for TC on collaborators’ local validation data, alongside 27% for ET and 16% for WT.The reported improvements were statistically significant for all three tumor sub-compartments.
  • Out-of-sample evaluation: 27% improvement was achieved for TC on complete out-of-sample data, alongside 15% for ET and 16% for WT.These comparisons evaluated the final consensus model against the public initial model on unseen-site data.
  • Interpretation: The final consensus model’s only difference from the public initial model was learning from the increased datasets contributed by all collaborators.The authors connect the resulting performance gains with access to larger and more diverse data.

Discussion

The study used federated learning across a globally distributed dataset to develop an accurate, generalizable glioblastoma boundary detector. The resulting model improved performance over the public initial model on both local validation and complete out-of-sample data, with relevance to treatment planning and further quantitative analyses.

  • Discussion: The final consensus model significantly outperformed the public initial model on collaborators’ local validation data and complete out-of-sample data.The authors attribute the improvement to access to larger and more diverse data through federated learning.
  • Discussion: The use case combined a multi-parametric, multi-class task with expert manual annotations and harmonized preprocessing across differing MRI acquisition settings.These requirements made the application larger and more complex than limited existing real-world federated learning studies.
  • Discussion: Glioblastoma boundary detection is relevant to neurosurgical and radiotherapy planning and is a first step toward downstream quantitative analyses.The authors propose that a continuous federated learning consortium could support routine practice and clinical trials.

Competing Interest Declaration

Intel-affiliated authors disclose potential competing interests related to OpenFL and brain tumor boundary-detection products. Intel may benefit commercially from products supporting the project.

  • Competing Interest Declaration: Intel-affiliated authors disclose potential competing interests because Intel may develop proprietary software related to the OpenFL project highlighted in this work.The disclosure also states that Intel may benefit from selling products supporting brain tumor boundary-detection models.
Loading 2204.10836v2…