Rxivist logo

Machine Learning for Large-Scale Quality Control of 3D Shape Models in Neuroimaging

By Dmitry Petrov, Boris A. Gutman, Shih-Hua (Julie) Yu, Theo G.M. van Erp, Jessica A. Turner, Lianne Schmaal, Dick Veltman, Lei Wang, Kathryn Alpert, Dmitry Isaev, Artemis Zavaliangos-Petropulu, Christopher RK Ching, Vince Calhoun, David Glahn, Ted Satterthwaite, Ole Andreas Andreasen, Stefan Borgwardt, Fleur Howells, Nynke Groenewold, Aristotle Voineskos, Joaquim Radua, Steven G Potkin, Benedicto Crespo-Facorro, Diana Tordesillas-Gutiérrez, Li Shen, Irina Lebedeva, Gianfranco Spalletta, Gary Donohoe, Peter V Kochunov, Pedro G.P. Rosa, Anthony James, Udo Dannlowski, Bernhard T. Baune, André Aleman, Ian H. Gotlib, Henrik Walter, Martin Walter, Jair C. Soares, Ruben C. Gur, N. Trung Doan, Ingrid Agartz, Lars T. Westlye, Fabienne Harrisberger, Anita Riecher-Rössler, Anne Uhlmann, DJ Stein, Erin W. Dickie, Edith Pomarol-Clotet, Paola Fuentes-Claramonte, Erick Jorge Canales-Rodríguez, Raymond Salvador, Alexander J. Huang, Roberto Roiz-Santiañez, Shan Cong, Alexander Tomyshev, Fabrizio Piras, Daniela Vecchio, Nerisa Banaj, Valentina Ciullo, Elliot Hong, Geraldo Busatto, Marcus V. Zanetti, Mauricio H. Serpa, Simon Cervenka, Sinead Kelly, Dominik Grotegerd, Matthew D. Sacchet, Ilya M. Veer, Meng Li, Mon-Ju Wu, Benson Irungu, Paul M. Thompson, for the ENIGMA consortium

Posted 21 Jul 2017
bioRxiv DOI: 10.1101/166496 (published DOI: 10.1007/978-3-319-67389-9_43)

As very large studies of complex neuroimaging phenotypes become more common, human quality assessment of MRI-derived data remains one of the last major bottlenecks. Few attempts have so far been made to address this issue with machine learning. In this work, we optimize predictive models of quality for meshes representing deep brain structure shapes. We use standard vertex-wise and global shape features computed homologously across 19 cohorts and over 7500 human-rated subjects, training kernelized Support Vector Machine and Gradient Boosted Decision Trees classifiers to detect meshes of failing quality. Our models generalize across datasets and diseases, reducing human workload by 30-70%, or equivalently hundreds of human rater hours for datasets of comparable size, with recall rates approaching inter-rater reliability.

Download data

  • Downloaded 529 times
  • Download rankings, all-time:
    • Site-wide: 29,461 out of 89,886
    • In neuroscience: 4,994 out of 15,993
  • Year to date:
    • Site-wide: 80,674 out of 89,886
  • Since beginning of last month:
    • Site-wide: 63,696 out of 89,886

Altmetric data

Downloads over time

Distribution of downloads per paper, site-wide


Sign up for the Rxivist weekly newsletter! (Click here for more details.)