Volumetric Multi-contrast Magnetic Resonance Imaging Biomarker for Predicting MRI-Guided Focused Ultrasound Treatment Outcomes

Magnetic resonance-guided focused ultrasound (MRgFUS) enables non-invasive tumor ablation, but real-time assessment of treatment efficacy remains challenging. Current clinical methods rely on delayed contrast-enhanced imaging or MRI derived thermal dose estimates, both of which are not adequate for predicting treatment efficacy. In this paper, we propose a deep learning model that utilizes efficient, volumetric multi-contrast MRI data acquired during MRgFUS ablation treatment to predict the MRgFUS induced non-perfused volume (NPV) measured 3 days after the ablation treatment, a strong indicator of tissue necrosis. An efficient, volumetric Multi Pathway Multi-Echo (MPME) sequence, an unbalanced multi configuration steady-state sequence, combined with MR thermometry-derived maximum temperature projection, achieved the highest predictive performance (Dice \(=\) 0.63, MDA \(=\) 3.46 mm). The proposed approach outperformed the current clinical standard of treatment-day NPV estimation using contrast-enhanced scans and further surpassed all combinations of conventional MR imaging, including T1-weighted/T2-weighted MRI, quantitative T1/T2 maps, and apparent diffusion coefficient maps, in a rabbit VX2 tumor model. This is a clinically significant result, as the efficient, volumetric MRI protocol required only 13 min of scan time, substantially less than the 66 min required for conventional 2D imaging protocols, while also offering large field of view, volumetric coverage. This efficient, gadolinium contrast-free protocol enables intra-procedural treatment guidance not available currently in MRgFUS treatment assessment.