Date of Award

Summer 8-16-2024

Level of Access Assigned by Author

Open-Access Thesis

Degree Name

Master of Science (MS)

Department

Biomedical Sciences

Advisor

Karissa Tilbury

Second Committee Member

Andre Khalil

Third Committee Member

Michael Mason

Abstract

Early detection of diseases and injuries is critical for many treatments to be effective in achieving positive patient outcomes; this early detection is aided by accessible and non-invasive diagnostic methods including biophotonics. Tools such as Spatial Frequency Domain Imaging (SFDI) and Diffuse Optical Spectroscopic Imaging (DOSI) are two clinically relevant and powerful biophotonics techniques for identification and monitoring of disease and injury in medical settings, due to their utility in imaging and diagnosis, but recognizing their full potential necessitates advances in photon simulations to improve accuracy. The utility of SFDI in particular is heavily limited by the accuracy and applicability scope of available lookup tables (LUTs) generated by Monte Carlo simulations, with a disproportionate impact on marginalized communities due to unaddressed effects of varying levels of skin melanin. Monte Carlo Extreme (MCX) is a software suite which allows a fuller utilization of modern computing power through GPU acceleration for running simulations, improving processing speed by up to more than 100x compared to Monte Carlo Command Line, a software suite for running Monte Carlo simulations already commonly in use. The Gardner method is a method using Monte Carlo simulation for generating LUTs dependent on a mathematical transform; the Fourier method is a potentially novel method of using Monte Carlo simulations for generating LUTs described in this paper which is more directly analogous to SFDI imaging processes which, we believe, could result in more accurate LUTs. In this work, MCX was utilized as a testbed to demonstrate the advantages and advances of the software with regard to rapid generation of LUTs and the potential of the Fourier method over the Gardner method. Specifically, we demonstrate the ability to spatially project patterns identical to those physically used in SFDI instrumentation. We benchmarked traditional Gardner-based Monte Carlo Simulation approaches in MCX as well as MCCL, comparing time requirements and accuracy of optical property determination for the Gardner method implemented in both software suites against an MCX Fourier illumination pattern approach. We found that MCX demonstrates speed improvements over MCCL ranging from 25x to upwards of 100x faster LUT generation speeds, and that the Fourier method showed marked advantages in accurate extraction of optical properties with a reduced rate of error compared to Gardner. The reduction of time required to generate various LUTs enables the ability to expand the range of LUTs to better encompass the role of melanin in various skin tones observed in clinical practice to improve accuracy of tissue chromophore extraction. Specifically, we found that failing to account for melanin concentrations associated with a broader range of skin-tones can result in upwards of 70% error in absorption coefficient extraction when using multi-layer LUTs with inaccurate skin optical property assumptions within the range of human skin tone variation. In summary, MCX-based Fourier-patterned photon simulations demonstrate promise in terms of reducing the computational burden of generating appropriate LUTs needed for implementing a broader and more accurate application of SFDI imaging in clinical practice.

Download

Share