Deep Learning-Based Glioma Segmentation of 2D Intraoperative Ultrasound Images: A Multicenter Study Using the Brain Tumor Intraoperative Ultrasound Database (BraTioUS)

Abstract

This study explores the use of artificial intelligence to improve the accuracy of intraoperative ultrasound (ioUS) imaging for glioma segmentation during neurosurgery. By training a deep learning model on data from multiple centers, this research demonstrates the potential for automated tumor delineation, despite challenges such as image noise and variability. The model was tested on independent datasets and showed strong performance overall, although external validation highlighted areas for improvement. Notably, the model demonstrated generalizability across ioUS systems from different scanner types and manufacturers, underscoring its robustness in diverse clinical settings. These findings emphasize the feasibility of using AI to enhance ioUS imaging, paving the way for more precise and efficient tumor resections in clinical practice.