Good news for data sharing: Defacing of MR scans using SimNIBS 4.0

Behnke, Larissa; Mizutani-Tiebel, Yuki; Chang, Kai-Yen; Thielscher, Axel; Bulubas, Lucia; Karali, Temmuz; Papazov, Boris; Kumpf, Ulrike; Stöcklein, Sophia ; Campana, Mattia; Soldini, Aldo; Dechantsreiter, Esther Maria Luisa; Tagnin, Lisa; Burkhardt, Gerrit; Takahashi, Shun; Padberg, Frank und Keeser, Daniel (2023): Good news for data sharing: Defacing of MR scans using SimNIBS 4.0. In: Brain Stimulation, Bd. 16, Nr. 1: S. 386

Abstract

Introduction: Transparency and reproducibility are important quality attributes in science. However, privacy and data security limit the public use of raw datasets in platforms like Open Science Framework. One possibility is to anonymize MRI data and make the facial profile (nose, eyes, mouth) unrecognizable using tools such as the Python package pydeface. For full anonymization all facial data is removed from the MRI scan. This reliably prevents facial reconstruction, which occurs when head-mesh files are generated by SimNIBS. Therefore, we aimed to investigate whether defaced and full-faced scans result in different electric-fields (e-field) when using SimNIBS. Methods: We simulated e-field distributions (n = 17) with SimNIBS (Version 4.0b0). Transcranial Magnet Stimulation was applied over the left DLPFC at an intensity of 80% of the resting-motor threshold. For each subject, the simulation was conducted with T1- and T1-T2-weighted scans (scan type) with i) defaced and ii) full-faced scans, respectively. Results: The ANOVA revealed no significant differences between defaced and full-faced scans for all e-field percentiles (99.9%, 99.0%, 95.0%) and a main-effect of scan type for the 99.9% e-field percentile. Post-hoc analysis showed higher e-field values for T1-weighted than for T1-T2 weighted scans for the 99.9%-percentile. Discussion: In summary, for e-field simulations with SimNIBS4.0 defaced scans do not result in different outcomes than full-faced scans. SimNIBS4.0 automatically substitutes the missing facial data by template data, which might contribute to the low differences. But different scan types might lead to diverging results in the highest e-field percentile. By ensuring the same e-field output, future studies with SimNIBS should be performed with defaced images, allowing the community to share scans and head-mesh files with the interested public in the spirit of the Open Science concept.

• Dokumententyp: Zeitschriftenartikel
• Fakultät: Medizin
• Themengebiete: 600 Technik, Medizin, angewandte Wissenschaften > 610 Medizin und Gesundheit
• ISSN: 1935861X
• Sprache: Englisch