Grants and Contributions:

Grant or Award spanning more than one fiscal year. (2017-2018 to 2022-2023)

The goal of low-dose rate (LDR) permanent-implant brachytherapy is to strategically place radionuclide seeds of sufficient radioactive strength near/inside the tumor volumes of interest (such as prostate and breast cancer) to damage tumor cells to an extent that they are either dead (necrotic) or not viable to reproduce (sterile), while minimizing damage to nearby healthy organs in the process (skin, urethra, rectum, etc.). It is, therefore, crucial to accurately estimate the 3D radiation dose distribution imparted by the seeds for assessment of the treatment quality and to correlate with prognosis.
MRI offers an excellent soft tissue contrast, which is a fundamental advantage over other medical imaging modalities (such as CT), for brachytherapy post-implant dosimetry. It also offers both anatomical and functional imaging capabilities that can be combined with dosimetric information to potentially provide powerful prognostic tools. Therefore, the goal of this program is to develop a comprehensive MRI based brachytherapy dosimetry protocol.
A major challenge, however, is the lack of MR signal from the radionuclide seeds, resulting in dark void appearance on the images. This makes the localization of the seeds difficult, which in turn, makes accurate dosimetric assessment impossible. In addition, the presence of calcifications (also having no MR signal, which are abundant in prostate and breast) adds to the uncertainty in seeds localization. No solution exists yet in distinguishing the seeds from calcifications.
The primary goal of this proposal, therefore, is to design and implement a robust approach that relies on a priori knowledge of the shape, geometry, and magnetic susceptibility of the radionuclide seeds, to reliably locate them with a positive contrast on MR images. Building on our promising preliminary work, this novel methodology will incorporate the design of a specific artificial neural network that benefits from the appearance of the seeds on various MR sequences. It will also incorporate a priori knowledge of the magnetic susceptibility-based distortions induced from the seeds as well as calcifications. A deep learning based framework is then proposed to accommodate inputs from various imaging sequences into an artificial neural network to automatically highlight the seeds and the calcifications, and separating them.
A secondary goal is to extract necessary radiologic parameters from the MR images to enable full heterogeneity-corrected dose calculations, empowered by Monte Carlo engine. That is, both electron density and effective atomic composition maps will be generated through novel quantitative acquisition and reconstruction techniques. Our preliminary results show excellent agreement in phantom tests.
Finally, dual energy CT (DECT) will be used to validate and quantify the accuracy of the seeds locations and the extracted radiologic parameters from MRI.