Computational Model of Optical Coherence Tomography in Lung Tissue: A Need for Speed!

Abstract

Little is understood about the detailed micro-mechanical properties of lung in-vivo because of the difficulty of imaging its heterogeneous structure. The goal is to improve imaging by better understanding light propagation through lung tissue so that better anatomical models and algorithms can be designed. Optical coherence tomography (OCT) is a commonly used technique to study the lung, so we simulated the OCT data with a Finite-Difference Time-Domain (FDTD) computer model. However, the algorithm took too long and left experimenters waiting. Also, the computational power scales rapidly for larger simulations and limits the size of the model being sampled. The goal was to optimize the current code to reduce the limits. First, MATLAB Parallel Toolbox was utilized, which enabled the simulation to achieve two-fold increase in depth in half the time. We then accessed Northeastern’s ‘Opportunity’ Cluster, which enables twelve CPUs to work in parallel, and obtained a three-fold increase in depth in one-tenth the time. This provided feasible analysis for transition towards the GPU, which was implemented next. However, due to the nature of the FDTD algorithm this proved difficult, and limited the gain at initial depth to one-third the time. The parallel implementation developed here for lungs could be applied to other tissues and imaging modalities as well. Changes in excitation and detection can be made to explore techniques such as confocal, multi-photon, and second-harmonic microscopy. Experimenters have already found the effects that elastin has on lung images due to its high refractive index along with its web-like geometry.