Automatic interstitial photodynamic therapy planning via convex optimization

Abstract

Finding a high-quality treatment plan is an essential, yet difficult, stage of Photodynamic therapy (PDT) as it will determine the therapeutic efficacy in eradicating malignant tumors. A high-quality plan is patient-specific, and provides clinicians with the number of fiber-based spherical diffusers, their powers, and their interstitial locations to deliver the required light dose to destroy the tumor while minimizing the damage to surrounding healthy tissues. In this work, we propose a general convex light source power allocation algorithm that, given light source locations, guarantees optimality of the resulting solution in minimizing the over/under-dosage of volumes of interest. Furthermore, we provide an efficient framework for source selection with concomitant power reallocation to achieve treatment plans with a clinically feasible number of sources and comparable quality. We demonstrate our algorithms on virtual test cases that model glioblastoma multiforme tumors, and evaluate the performance of four different photosensitizers with different activation wavelengths and specific tissue uptake ratios. Results show an average reduction of the damage to organs-at-risk (OAR) by 29% to 31% with comparable runtime to existing power allocation techniques.

Optimizing interstitial photodynamic therapy with custom cylindrical diffusers

“…Our testing methodology in this work is based on . We run all our tests on a 3D tetrahedral mesh of a segmented brain atlas .…”

“…PDT‐SPACE with point sources was able to achieve an average of 30% less damage to the surroundings of the tumor in virtual glioblastoma multiforme cases with the same tumor coverage when compared to the Cimmino algorithm, a widely used linear feasibility algorithm for power allocation . However, the number of sources required to achieve a good coverage of the tumor in was high. This high number of sources makes the surgery more complex and time‐consuming, and results in more mechanical perforations through both healthy and diseased tissues .…”

“…To create an optimized plan, we need to determine the minimal number of light sources required for the treatment, along with their locations and power allocation. Recently, we introduced a PDT Source Power Allocation using Convex optimization Engine, or PDT‐SPACE . This tool allocates power to a fixed set of sources, as the name suggests, using convex optimization , which guarantees an optimal solution to minimize the damage to an OAR.…”

“…. where the matrix A is used to set the power constraints on certain combinations of sources, that is, it can be used to supply per‐source power constraints and/or a total power limit for some (or all) sources. p max is a column vector of size equal to the number of constraints (rows) in the matrix A , each element of which is the maximum power limit for the corresponding constraint .…”

High-performance, robustly verified Monte Carlo simulation with FullMonte

“…Since FullMonte is open-source with a modular design, advanced users may also customize and/or link FullMonte into their own C++ code, as we have done to apply FullMonte to convex optimization of PDT source power allocation and placement. 11 Below, we provide details of each of the phases of an in-silico experiment with FullMonte.…”