Automated Identification of Mod-Sev TBI Lesions 2026¶
This challenge is part of the MICCAI 2026 challenges (https://conferences.miccai.org/2026/en/challenges.asp) so participants are expected to submit a 6-8 page paper on their algorithms and results. This paper will be due July 23rd 2026. The in-person challenge event will take place in Strasbourg, France on October 1st 2026. No travel support will be provided by challenge organizers, but the top 3 performing teams will receive a cash prize - 500USD for 1st place, 250USD for 2nd, and USD100 for 3rd - along with brain trophies!
Data access¶
Thank you for your interest in the AIMS-TBI segmentation challenge! Please start by requesting access to the training data: https://forms.gle/GHEEYkeUojGFfgM58
You will need to agree to the data use requirements¶
Additional details available here: https://github.com/adionicas/AIMS-TBI-challenge
Background¶
Moderate to severe TBI (msTBI) results from external forces causing rapid brain movement, triggering complex pathophysiological changes. Primary injuries — including hematomas, hemorrhages, and contusions — initiate cascading secondary injuries such as gliosis and encephalomalacia, which can cause life-threatening complications requiring acute surgical intervention. Each of these processes contributes to structural brain deformation, and the unique accumulation of changes across patients produces the extremely heterogeneous lesion profiles that characterize msTBI. Unlike lesions in stroke, multiple sclerosis, or brain tumors, msTBI lesions can be focal or diffuse, vary in size, number, and laterality, extend across multiple tissue types, and occur bilaterally in homologous regions. This complexity complicates image registration, normalization, and brain parcellation — introducing both local and global errors in standard neuroimaging pipelines. Current approaches to managing lesions in msTBI processing are inadequate: ignoring lesions produces unreliable findings, excluding patients with large lesions limits generalizability, and manual segmentation is too time-consuming to scale. Existing automated tools either require multiple image types — T1, T2, FLAIR, gradient echo, and proton density — or are limited to CT, restricting their use in large multi-site MRI consortia. What is needed is an accurate, automated lesion segmentation algorithm trained on large, multi-cohort MRI datasets that can run on commonly acquired sequences.
The AIMS-TBI challenge leverages data from the ENIGMA Consortium's Traumatic Brain Injury working group, specifically the Pediatric and Adult msTBI subgroups. The challenge focuses on T1-weighted (T1w) MRI exclusively — the most common sequence across ENIGMA TBI sites and the most consistent in acquisition parameters across sites compared to other modalities such as diffusion MRI. The goal is to develop algorithms that accurately detect and segment 3D lesions visible on T1w MRI resulting from TBI, including contusions, hemorrhage, hematoma, encephalomalacia, gliosis, white matter lesions, and surgical drainage tracts.
The inaugural AIMS-TBI Challenge was held at the 2024 MICCAI conference and it was held again in 2025. The final AIMS-TBI 2025 dataset comprised a total of 892 images, including 553 images for training, 100 for validation, and 239 for testing. In the validation phase of the challenge, 15 teams uploaded 60 entries, with 9 teams submitting final models in the test stage. The best performing model achieved an average Dice score of 0.637 overall, and the highest Dice for the lesion-only files being 0.515. This result represented an improvement over the first year, yet also highlighted substantial room for continued development. This year we will separate the tasks of segmentation and detection, identifying which subjects are lesion-free in the training data and focusing the performance metrics on the cases with lesions and having dual leaderboards for each task. Additionally, this year we will make multimodal MRI data available for teams to use in training their algorithms.
TASKS¶
1. Detect lesions in T1-weighted MRI data from moderate-severe TBI
2. Segment lesions in T1-weighted MRI data from moderate-severe TBI
Inputs
- T1-weighted MRI images
- Tabular demographic and clinical data
Outputs
- Binary TBI lesion segmentation mask.
SCHEDULE¶
- May 15: Training data released
- May 30: Validation phase opens
- July 1: Final model submission opens
- July 23rd: Deadline for submission dockers