BraTioUS: A multicenter dataset of baseline intraoperative brain tumor ultrasound images

TL;DR

BraTioUS is a large, diverse dataset of brain tumor ultrasound images from multiple hospitals, designed to improve AI tools for glioma surgery.

Contribution

The dataset introduces a large-scale, multicenter, and publicly available collection of intraoperative ultrasound images with expert tumor segmentations.

Findings

BraTioUS contains 1669 2D ioUS images from 142 glioma patients across six hospitals.

The dataset includes expert-annotated tumor segmentations for each image.

It addresses limitations in existing datasets by offering diversity in hardware, protocols, and glioma types.

Abstract

The BraTioUS (Brain Tumor Intraoperative Ultrasound) dataset [1] is a large-scale, multicenter, and publicly available collection of intraoperative ultrasound (ioUS) images acquired during glioma surgeries. Created through an international collaboration among six hospitals across five countries, BraTioUS comprises 1669 B-mode 2D ioUS images from 142 glioma patients collected between 2018 and 2023 using various ultrasound systems and acquisition protocols. It also includes masks supervised by experts of tumor segmentation from every ioUS image.

BraTioUS addresses several limitations found in existing public datasets, such as lack of diversity in acquisition hardware, imaging protocols, and glioma types. The primary objective of this dataset is to be publicly available and accessible for the training and validation of machine learning models aimed at improving the interpretation and use of ioUS. The dataset’s scale, quality, and heterogeneity make it a valuable resource for training and validating AI tools aimed at improving intraoperative decision-making and patient outcomes in glioma surgery.

Figures