Summary/Abstract While improved early detection methods and treatments have reduced breast cancer mortality, a sizable portion of patients remains overdiagnosed and overtreated, warranting the development of more conservative breast cancer treatments. Magnetic resonance guided focused ultrasound (MRgFUS) is one of the most attractive, emerging procedures for breast cancer as it can safely and efficaciously treat localized breast tumors non-invasively. Currently, clinical MRgFUS ablation treatments are assessed with MRI metrics that primarily quantify thermal and vascular effects. While there is evident MR sensitivity for tissue changes induced by MRgFUS thermal ablation, no single metric or combination of metrics have demonstrated adequate accuracy in predicting tissue viability during or immediately post-MRgFUS ablation treatment. In addition, the use of gadolinium contrast agent-based assessment techniques precludes further ablation treatment if positive tumor margins are suspected. This work proposes to address this critical unmet need through developing a deep neural network non-contrast imaging biomarker that would provide an immediate and accurate assessment of tissue viability and could be applied repeatedly for an iterative assessment of tissue viability during the MRgFUS ablation procedure, assuring complete non-invasive tumor treatment. This objective will be accomplished with three specific aims. Aim 1: Develop and validate a 3D multiparametric MRI protocol for efficient acquisition of qualitative and quantitative MR images in the breast MRgFUS therapeutic environment. Aim 2: Develop, train and validate a deep neural network biomarker for predicting tissue viability in a tumor model during MRgFUS ablation treatments. Aim 3: Integrate the tissue viability biomarker in an existing breast MRgFUS ablation clinical workflow and demonstrate complete treatment volume ablation using the non-contrast, deep neural network biomarker as the treatment assessment metric. We have developed an innovative, volumetric histopathology diffeomorphic registration procedure that allows the voxel-wise comparison of in vivo MR images to histopathological data, providing the gold-standard labeled data set needed to develop this imaging biomarker. Training and validation of the imaging biomarker will be performed in preclinical models designed to allow immediate generalizability and translation to ongoing clinical trials. This imaging biomarker will provide accurate assessment of tissue viability during MRgFUS ablation treatments, revolutionizing minimally invasive breast cancer treatments and directly addressing the critical issue of overtreatment.

Narrative The current treatment paradigms for minimally invasive ablative breast cancer therapies do not ensure complete tumor ablation or allow for immediate and accurate assessment of tissue viability. Although several MRI contrasts are sensitive to thermal ablation, no individual or combined MR contrasts have been able to accurately predict tissue viability immediately after magnetic resonance guided focused ultrasound ablation. This proposal will develop a non-contrast, deep neural network imaging biomarker that will provide both an accurate assessment of tissue viability and a treatment outcome metric that will allow for iterative MRgFUS ablation assessment, ensuring efficacy of magnetic resonance guided focused ultrasound ablation treatments for breast cancer.