Research – Godinez Lab

Machine Learning in MRI image analysis

To characterize the classification performance of a Bayesian Additive Regression Trees (BART) model, built from MR radiomic features, and compare it to the readings of a musculoskeletal radiologist, in classifying atypical lipomatous tumors (ALTs) from simple lipomas.

A cohort of 423 patients was included and 1132 radiomic features were extracted from each MR study. The BART model had an accuracy, sensitivity, and specificity of 77.07% (72.76%-80.99%), 77.67% (71.36%-83.16%), and 76.50% (70.28%-81.97%), respectively, when utilizing all predictors and aggregating training and testing data from all the cohorts, approximating the human reader 78.72% (74.51%-82.53%), 76.21% (69.80%-81.85%), and 81.11% (75.25%-86.09%), respectively. In the external validation, the average AUC value across cohorts between the BART model and the human reader differed by 0.04 AUC points. From the receiver operating characteristic curve, the AUC was calculated to be 84.72% (81.00%-88.50%) and 84.74% (81.00%-88.50%) for the BART and human reader, respectively.

This study demonstrated that the BART model can distinguish ALT from lipoma with diagnostic performance comparable to an experienced human observer.

Chemical Exchange Saturation Transfer

The CEST signal originates from labile hydrogen protons attached to tumor-derived molecules such as amino acids, creatine, sugars, and other proteins that are upregulated in tumor cells. These molecules are present in concentrations too low for conventional MRI to detect directly; instead, CEST MRI indirectly measures their signal through a magnetic property called magnetization transfer, which includes the chemical exchange pathway. Many parameters, such as pH, solute concentration, and chemical exchange rate, can be derived from CEST using the Bloch-McConnell model. This provides a rich source of molecular probes. Our goal is to translate CEST from preclinical to clinical utility. We are developoing software to perform the complicated CEST signal analysis and data visualization that can streamline the evaluation of clinical CEST studies. We are exploring the application of CEST in imaging human joints and in soft tissue cancers.

Interventional MRI in Cancer Therapy Delivery

Endovascular selective inter-arterial infusion (ESIAI) is a technique that introduces a drug with endovascular devices via the main arterial blood vessel feeding the tumor. ESIAI is gaining interest as a selective delivery solution in GBM. ESIAI in its current form requires X-ray fluoroscopy to visualize the device and MRI to image the tumor. Transferring patients between modalities and MR incompatible endovascular devices limit its practicality and use. MRI is superior to X-ray for guiding ESIAI, as it directly visualizes brain parenchyma perfusion and facilitates predictable drug targeting and offers a variety of tissue contrasts for cancer imaging. Device visualization with MRI would improve the workflow of ESIAI and the accuracy of co- locating the device and tumor which is critical to ensure target coverage while minimizing collateral damage to benign tissue. However, the tips of metallic endovascular devices have poor conspicuity and can heat causing irreversible tissue damage [26]. Redesigning the device with non-metallic materials has been attempted, resulting in a reduction in mechanical performance. I propose a method that converts guidewires and microcatheters into active transmit and receive devices (ATRD) without altering the device, to achieve MRI compatibility and visualization of standard devices, so that moving patients between modalities is not needed while engaging all the imaging benefits of MRI. This method can be combined with parallel transmit (pTx) MRI, which provides precise control over the excitation field that is responsible for the heating effects, and it can also be used to bolster a device’s conspicuity. I have researched this method for cardiac applications, and I estimate it can work well in the brain. Having the ability to guide ESIAI with MRI while still performing assessment protocols is needed to streamline the procedure. A key assessment in cancer management is evaluating the tumor extent which is associated with the BBB status. Although the management of GBM may be a combination of delivery strategy, targeting agents, and adaptive treatments, in this work I will focus on advancing the delivery strategy with novel interventional MR imaging tools.

Modeling of guidewire RF behavior inside MRI

Metallic Endovascular devices can be unsafe when used during MR guided procedures. A potential solution to the heating problem might be to use a local coil instead of a body coil to excite the imaging volume. In this work we present a Monte Carlo analysis using many random guidewire trajectories and the transfer function to estimate the scatter electric field at the guidewire tip imbedded in a heterogenous dielectric.

PET Insert for 7T MRI

Simultaneous preclinical PET/MRI is well suited to applications which require a dynamic process to be imaged both with PET and MRI. Parametric measurements from PET and functional parameters from MRI can then be used together to gain a better understanding of biological processes. We have designed and built a preclinical PET/MRI insert with sufficiently high sensitivity to image such dynamic processes in mice and rat brains. The PET system incorporates a silicon photomultiplier based detector which is itself fully MR compatible. All data transferred out from the scanner is transmitted optically, to reduce RF leakage across the faraday cage. In this work we assess the impact of the PET insert on image quality of a 7 Tesla animal MR scanner.