Join

BONBID-HIE

Lesion Segmentation and Outcome Prediction

Data

The complete data is organized in the format shown below. 

BONBID-HIE provides, per patient for MICCAI 2024 Challenge lesion prediction: 



  • 1ADC_ss: skull stripped Apparent Diffusion Coefficient (ADC) map. 
  • 2Z_ADC: ZADC map.
  • 3LABEL: expert lesion annotations.
  • 2-year Outcome.

For  data descriptions of lesion segmentation, please read and cite our paper.  

Rina Bao, Ya'nan Song, Sara V. Bates, Rebecca J. Weiss, Anna N. Foster, Camilo Jaimes Cobos, Susan Sotardi, Yue Zhang, Randy L. Gollub, P. Ellen Grant, Yangming Ou "BOston Neonatal Brain Injury Dataset for Hypoxic Ischemic Encephalopathy (BONBID-HIE): Part I. MRI and Manual Lesion Annotation". bioRxiv 2023.06.30.546841; doi: https://doi.org/10.1101/2023.06.30.546841

Data descriptions of outcome labeling:

https://github.com/baorina/BONBID-HIE-MICCAI-MICCAI2024/blob/main/MICCAI2024-BONBID-part2.pdf


Training Set


To be released 8/15/2024  9/1/2024  9/6/2024

Track1: HIE Lesion Segmentation

Input: ADC, ZADC

Output: Lesion segmentation map


labeled MGH data. https://zenodo.org/records/10602767

Please download version 3; it is fully accessible and does not require access requests. BONBID2023_HIE Test is encrypted with a password. Since this set is intended for testing, the password will be made public after the challenge.

unlabeled BCH data https://zenodo.org/records/13690270

In this task, participants are tasked with training models to segment HIE lesions using the provided MGH and BCH datasets. You are encouraged to explore different learning approaches, including supervised, semi-supervised, or unsupervised learning methods. There are no restrictions on the choice of methodology, giving you the freedom to innovate and experiment with the data to develop the most effective segmentation models. Your objective is to leverage these datasets to achieve the best possible performance in accurately segmenting HIE lesions, advancing the challenge through your innovative solutions.

Track 2: HIE 2-Year Outcome Prediction Using MGH and BCH Data (Binary Classification)

Input: ADC, ZADC

Output: Outcome label


Outcome label  https://zenodo.org/records/13690270

For the description of data, please find it here

In this track, participants are challenged to train models for predicting 2-year outcomes of HIE patients using the provided MGH and BCH MRI datasets. The task involves developing models that can accurately classify the outcome as a binary value: 1 for adverse outcome and 0 for normal outcome. Participants are encouraged to explore various modeling techniques and leverage the data to create robust predictive models that can effectively distinguish between normal and adverse outcomes at the 2-year mark.


Validation Set

Validation data 

N=4 cases for both tracks. (Cases are from training samples)
The small validation set is only used for participating teams to do a sanity check of algorithm dockers. The performance won't be used to rank teams.


Testing Set

Testing data (hidden set)

Track 1: N=44 cases

The algorithm dockers submitted will be run on the test set. The final performance and ranks will be evaluated on the test set. 

Track 2: N=53 Cases

The algorithm dockers submitted will be run on the test set. The final performance and ranks will be evaluated on the test set.