BONBID-HIE 2023: Lesion Segmentation Challenge in BOston Neonatal Brain Injury Data for Hypoxic Ischemic...

Area of Science:

• Medical Imaging
• Neonatal Neurology
• Artificial Intelligence in Medicine

Background:

• Hypoxic Ischemic Encephalopathy (HIE) affects 1-5/1000 neonates, causing brain dysfunction.
• Accurate segmentation of HIE lesions in neonatal MRI is vital for prognosis and treatment evaluation.
• Limited annotated data has hindered the development of automated HIE lesion segmentation algorithms.

Purpose of the Study:

• To address the data scarcity challenge in HIE lesion segmentation.
• To organize the first BONBID-HIE challenge using diffusion MRI data (ADC maps).
• To evaluate the performance of machine learning-based segmentation algorithms for HIE lesions.

Main Methods:

• Organized the BONBID-HIE challenge in conjunction with MICCAI 2023.
• Collected and utilized diffusion MRI data (Apparent Diffusion Coefficient maps) for HIE lesion segmentation.
• Received 14 submissions employing various automated machine learning segmentation algorithms.

Main Results:

• Facilitated an in-depth evaluation of current HIE lesion segmentation technologies.
• Provided insights into the capabilities and limitations of submitted algorithms.
• Identified key areas for future research and development in the field.

Conclusions:

• The BONBID-HIE challenge successfully spurred advancements in automated HIE lesion segmentation.
• The challenge highlighted the need for continued research to overcome persistent hurdles.
• An annotated dataset, algorithm dockers, and evaluation codes are publicly available to foster further research.