Non-invasive neurosurgical planning with Random Matrix Theory MRI

PROJECT SUMMARY This application is intended for I-CORPs at NIH. Summary of the associated NIH NCI STTR Phase I About 23,830 people in the US are diagnosed per year with primary malignant brain tumors, and 200,000- 300,000 with metastatic brain tumors (10-30% of all cancers). Maximizing surgical resection of tumor is a major predictor of survival, but must be balanced against the risk of injuring eloquent white matter and cortical regions. To improve outcomes, the unmet need is to radically increase quality of noninvasive preoperative brain mapping. As the brain mapping gold standard, MRI offers unique soft-tissue contrast, anatomical and functional information of the brain, yet is inherently signal-to-noise ratio (SNR)-starved. The majority of brain mapping relies on diffusion (dMRI) and functional (fMRI), which are both especially severely limited by SNR. The MRI signal can be increased with higher-field; however, scanner prices scale with the field strength: 1.5T ~ $1.5M, 7T ~ $7M, as do installation and service costs. Since 90% of the MRIs in the US are 1.5T or below, it appears that the majority of hospitals cannot justify or afford high field MRI. SNR increase by the signal averaging is impractical from the scan time perspective, as brain tumor patients rarely tolerate scan times above 45 min. Our company, Microstructure Imaging (MICSI), is an award-winning New York University (NYU) spinoff that offers a software-as-a-service for medical image processing. Our product dramatically enhances the SNR of MRI brain mapping, which translates into increased resolution, image quality, sensitivity and specificity. Here we employ random matrix theory (RMT) to achieve an order-of-magnitude gain in SNR purely in software at the image reconstruction level, by utilizing the information across multiple radiofrequency coils and MRI contrasts within a single protocol. Our overarching goal is to optimize our RMT/MP-PCA image reconstruction algorithm for the clinical translation in brain mapping preoperative studies. Our Specific Aims are: Aim 1: Enabling lower field / higher resolution. We will develop and evaluate a multimodal (dMRI/fMRI) RMT denoising and reconstruction protocol in 6 volunteers on 1.5T and 3T with different image resolutions, and retrospectively in 30 preoperative brain mapping MRI patients. This data will be used to justify prospectively altering clinical MRI protocols during the anticipated Phase II of the STTR. Aim 2: Clinical feasibility study. 15 minutes of additional scan time for dMRI and 2 fMRI tasks at 1.2 mm isotropic resolution will be prospectively added to 10 brain mapping cases at 3T. The image quality with and without denoising will be assessed quantitatively, and qualitatively by radiologists and neurosurgeons. While the Phase-I STTR will optimize RMT in preoperative planning for brain tumors, in the future we will optimize protocols for any tumor type or location by joint RMT reconstruction of variety of MRI modalities (perfusion, T1/T2, dMRI, fMRI) to help them denoise each other and maximize the overall information content. RMT image reconstruction will open MRI to the developing world by bringing high-field quality to inexpensive low-field MRI.PROJECT NARRATIVE This application is intended for I-CORPs at NIH. Summary of the associated NIH NCI STTR Phase I A modern MRI scanner is a multi-million-dollar machine, which has become a clinical gold-standard for non- invasive diagnostics and presurgical planning of brain tumors; however, its overarching technological challenge is the signal-to-noise ratio (SNR), ultimately limiting the image resolution and quality in the clinic. To improve diagnostic image quality and increase the resolution, the SNR must be radically increased, which currently entails spending an additional 1-4 million dollars on higher-field scanners, with incremental outcomes. This proposal demonstrates that our innovative denoising and image reconstruction algorithm, based on random matrix theory and implemented purely in software, increases SNR by an order-of-magnitude in clinically relevant structural and functional MRI acquisitions, matching or exceeding the quality of costly hardware investments, and can greatly improve the quality of non-invasive neurosurgical planning for brain tumors.