Navigating the Future of Transcranial Focused Ultrasound: AI-Driven Innovations

Abstract

Objectives: We aim to present AI-driven innovations in transcranial focused ultrasound (tFUS) that enhance precision and efficacy in non-invasive therapeutics for neurological and psychiatric disorders.

Methods: Two AI-driven methods were employed: (1) Synthetic CT (sCT) generation from T1-weighted MRI using a 3D conditional generative adversarial network (3D-cGAN) was developed and compared to real CT (rCT) for tFUS acoustic simulations. (2) A real-time acoustic simulation framework using a 3D-cGAN was integrated with conventional image-guided navigation, tested for computational efficiency and accuracy in predicting intracranial acoustic fields.

Results: The sCT generation from MRI achieved a mean absolute error (MAE) of 280.25±24.02 HU (skull), with a dice coefficient similarity (DSC) of 0.88±0.02 (skull). Comparisons between rCT and sCT in acoustic simulations showed less than 4% difference in peak acoustic pressure and less than 1 mm difference in focal point location. The 3D-cGAN-based simulation-guided navigation (SGN) system achieved a frame rate of 5 Hz (0.2 seconds per frame) with errors of 6.8% in peak intracranial pressure and 5.3 mm in acoustic focus position. Experimental validation showed 4.5% peak intracranial pressure error and 6.6 mm focus position error.

Conclusions: AI-driven innovations in synthetic CT generation and real-time acoustic simulation enhance tFUS precision and safety, paving the way for more effective non-invasive brain treatments.