ENHANCING GLIOBLASTOMA THERAPY: COMBINING DEEP LEARNING SEGMENTATION WITH ROBOTIC HISTOTRIPSY

Abstract

Glioblastoma multiforme (GBM) is the most aggressive and prevalent adult brain cancer, with a median survival of 14 months due to its rapid growth, invasiveness, and resistance to conventional treatments. Current challenges include incomplete surgical removal caused by GBM’s infiltrative nature and limitations in intraoperative visualization techniques, posing significant risks to surrounding healthy tissue. Minimally invasive technologies like histotripsy, which uses focused ultrasound for mechanical tumor ablation, offer promising alternatives but require precise tumor delineation and targeting. This research addresses these challenges by developing an AI-driven framework for automatic tumor segmentation and robotic control for accurate histotripsy delivery. Testing on ex vivo and in vivo models demonstrated minor under-treatment. Similarly, in mouse models injected with GL261 cells, a manageable over-treatment was observed, which can be adjusted due to the flexible design of the developed algorithm. Additionally, applying a deep learning model achieved high real-time tumor segmentation performance, significantly enhancing both accuracy and efficiency.