Registration is open through August 15th for the MIDI-B Challenge – the NCI Medical Image De-Identification Benchmark Challenge. This is for developers (independent, academic, or industrial) who want to assess their automated image de-identification algorithms and software tools’ performance against a dataset of DICOM (Digital Imaging and Communications in Medicine) images that will be made available to challenge participants.

The top three ranked teams from this challenge will receive invitations to present virtually about their algorithms and results at the 27th International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI) in October. These teams will also collaborate on the development of a manuscript for a peer-reviewed publication. 

A fundamental requirement for sharing medical images through publicly accessible data commons and repositories is removing patient identifiers, while following best practices to preserve the research value of the data. The DICOM dataset for this benchmark challenge contains synthetic identifiers for participants to work with, and no actual identifiable patient data will be made available in this select dataset. 

As part of the preliminary training, participants can assess their image de-identification models with a publicly available subset of the MIDI-B challenge data, through the NCI’s The Cancer Imaging Archive (Pseudo-PHI-DICOM-Data), as well as through the CRDC’s Imaging Data Commons (Pseudo-PHI-DICOM-Data). Participants can also train on their own DICOM datasets, but two larger  DICOM datasets, one for the validation phase and one for the test phase, will be available by registering and participating in the challenge.

Learn more and register for the MIDI-B Challenge. 

Important dates: 

• Registration – open through August 15
• Training phase – open through August 3
• Validation phase – August 4-30
• Test phase – September 4-6
• MIDI-B virtual session at MICCAI 2024: October 10

NCI CBIIT leads the organization of the MIDI-B Challenge in collaboration with the NCI’s National Heart, Lung, and Blood Institute, Sage Bionetworks, Frederick National Laboratory, University of Arkansas for Medical Sciences, Ellumen, Inc., PixelMed Publishing, Google, and Deloitte.