bencro wrote: Eoganacht ... Wow! Another very pertinent catch. Thanks for sharing.
Clearly focused on the brain cancer indication.
And most probably in line with this
Nov. 8 press release statement:
... Optimization of the dose of TLD-1433 to use in a Phase II clinical study, coupled with optimization of both the TLC-3200 and TLC-3400 DFOC technology, currently underway by the Company, to increase ease of manufacture and optimize uniformity of light distribution, will be important to successfully achieve a primary efficacy outcome measure. Below, is your excerpts. I only added
in green, additional passages that I found very interesting, fro a much global perspectives, as I also find
your yellow ones very appropriate.
For example,
Jeff Cassidy is back. He has the one behind our current TLC-3200/TLC-3400 (DFOC) version. He was under the supervision of Prof. Betz and Dr. Lilge.
William Kingsford is a new name. From Univ. of B-C, Dept. of Mathematics.
Theralase and IBM are also part of the fundings (grants).
Again, amazing find from you. Thanks for sharing. They clearly know what it takes to get there.
Optimizing Light Delivery In Photodynamic Therapy Biomedical Optics Express Vol. 9, Issue 2, pp. 898-920 (2018) Abdul-Amir Yassine, William Kingsford, Yiwen Xu, Jeffrey Cassidy, Lothar Lilge, and Vaughn Betz 1. Introduction and prior work
Photodynamic therapy (PDT) is the light activation of photosensitizers, inducing the production of oxygen radicals with the aim to destroy particular tissues in situ. The process starts by administering a photosensitizer to the patient, which will preferentially accumulate in the target tissue, followed after a photosensitizer-specific time delay with the irradiation of the treatment region with photosensitizer-specific visible or near-infrared light. The photosensitizer absorbs the light and initiates a series of photochemical reactions that generate cytotoxic products which lead to the desired treatment effects [1]. PDT has shown high efficacy in treating several superficial tumors such as skin cancer [2] and esophagus tumors [3] through surface illumination techniques. A more efficient illumination scheme that can target deep tissues is interstitial light delivery, in which fiber-based spherical light diffusers are placed directly into the target tissues [4]. However, several pre-clinical studies have shown that the treatment response with interstitial delivery is strongly dependent on the light dose distribution, which is itself dependent on the optical properties of the patient’s tissues and the placement of the light sources [5–7]. Therefore, the success and efficacy of interstitial photodynamic therapy (iPDT) strongly depends on planning the photon source placement based on a patient’s particular anatomy of the target and the surrounding tissues.
An iPDT treatment plan represents the proposed flow of the upcoming therapy. It specifies in particular the photosensitizer drug to use, its activation wavelength, the treatment time, and the number of light sources to use, their positions and the power for each light source. The success of PDT is contingent on attaining a sufficiently high cytotoxic dose in the target where the effective dose can be given by any of the existing PDT dose descriptions such as the photodynamic threshold model [8]. A perfect plan would result in a photon source distribution that leads to the complete destruction of the tissue in the target volume without affecting any surrounding healthy tissues, particularly organs at risk (OAR), i.e. healthy tissues that can tolerate only a limited amount of PDT dose without being harmed. Such a perfect plan usually does not exist, and clinicians face several challenges that hinder the establishment of a high-quality plan to destroy the target volume with minimal damage to the OAR.
To create an optimized treatment plan, we first need to accurately evaluate the plan by modeling the light distribution inside the tissues, the forward problem. For a given photon source, the light distribution, or fluence rate [mW/cm2], will vary among patients as it is dependent on the patient’s tissue optical properties and anatomy. The radiative transfer equation (RTE) which governs this process is generally not solvable for most realistic media of interest [1] and thus approximation methods are needed. Previous approaches to PDT treatment planning primarily used two methods. The simplest approach assumed an empirically determined radius around each source, in which cells are destroyed, and added sources with this kill-radius to cover the target tissue, while leaving other tissues unharmed [9,10]. A more sophisticated method is to use a diffusion approximation of light transmission in combination with either homogeneous [4] or heterogeneous optical properties [11–13], while applying a fluence [Jcm−2] threshold above which a tissue is destroyed. The main weakness of both approaches is the failure to accurately model the reflection and refraction of light at tissue boundaries with different optical properties, which often occur in biological systems [14]. They also do not take into account direction-dependent scattering and assume isotropic scattering events, which is not the case within a few mean-free paths from the photon sources, boundaries or photon sinks. Other discrete and continuous analytical solutions have been studied recently [15], and were found to be accurate only within homogeneous tissues. To accurately model tissue inhomogeneities and scattering events, we have used the Monte Carlo photon simulator FullMonte [16], one of the fastest accurate simulators of dose distribution published to date. By combining the light distribution computed by FullMonte with the relative photosensitizer concentration and the various tissues uptake, we can compute the PDT response throughout the volume of interest.
The second component of treatment planning is determining the minimal number of light sources required, their locations, and their power allocation – the inverse problem – which remains a major challenge in iPDT planning. This step is of utmost importance as it affects the treatment efficacy and is known to be highly correlated to the damage caused to the OAR [1]. Finding a high-quality treatment plan generally involves a sub-problem of finding a good power allocation among the light sources. Several attempts on this problem have been investigated in the literature, and the best allocation algorithm to date is the Cimmino linear feasibility algorithm [4,12,17]. A more general problem to achieve better-quality plans is to optimize the sources’ locations along with their power allocation. This problem is highly non-linear, and there is no technique to our knowledge that finds the optimal solution in a reasonable time for clinical praxis. Several attempts at treatment planning varied the possible source positions with a variety of sub-optimal perturbation algorithms [4,11,18]. Another approach used for source position optimization is Powell’s method for nonlinear optimization, which finds a local minimum of an optimization cost function, at the risk of potentially missing the global minimum [9, 10]. The implementation of Powell’s method relies on a simplified dichotomous PDT dose model which assumes uniform fluence and complete tissue destruction within a certain radius, while tissues outside are unaffected. Another significant downside of most of the proposed source placement optimization techniques is to assume homogeneous tissues, such as the prostate, where the vascular perfusion tree supports the confinement of the light in the gland limiting its exposure to the surrounding OAR. Additionally, over-treatment within the prostatic gland is not necessarily considered a major drawback [4,11,19,20]. Applying those approaches to tumors inside optically heterogeneous tissues, such as the brain, can yield low-quality plans that can not achieve the desired effect without potentially affecting eloquent areas of the brain.
In this work, we propose a fast and effective iPDT planning optimization work flow for most tissues, here evaluated on synthetic brain tumors. The proposed implementation operates on 3D meshes of segmented MRI or CT scan images of the target volume. Moreover, it assumes that the photosensitizer drug to use, its tissue uptake, and the tissue optical properties are all known to the clinician. The fluence dose threshold for each tissue can be determined based on that tissue’s response to the cytotoxic reactions due to the photosensitizer being used. Finally, the proposed approach requires a light propagation simulator to determine the fluence rate distribution for each source placement. As mentioned earlier, we use FullMonte [16] for this purpose. Given the aforementioned inputs,
we propose a cost function that can be efficiently minimized to create a power allocation for a set of fixed light sources. The proposed power optimization approach generalizes the implicit least-squares cost function of the Cimmino algorithm [21] to the broader space of convex optimization problems, in this case to a linear program (LP, an optimization model with requirements expressed by linear relationships [22]). It has the benefit of allowing the use of a wide class of fast algorithms that are guaranteed to converge to a global optimum, along with the flexibility in adding constraints (such as total or per-source power limits). The latter can allow clinically relevant factors to be taken into account, such as minimizing treatment duration, which for fixed target dose, would imply higher irradiation power, potentially at the cost of slightly more damage to healthy tissues. We then leverage the robustness, speed and guaranteed optimality of this power allocation program to select and iteratively refine the light source locations to further optimize the treatment plan. ....
6. Discussion
In the first part of this study, we have introduced a convex and linear power allocation program for interstitial PDT planning. The program returns an optimal power allocation solution across a certain number of light sources that minimizes the damage to the OAR. We have tested our program on synthetic brain tumors modeling real GBM images, and have quantified the quality of our plans using different photosensitizers activated at different wavelengths with different tissues’ responses (uptake ratios, optical properties and photosensitizer concentration). We have validated how the program can be controlled to increase the coverage of the tumor by tuning the importance weight of that tissue, or how the resulting plan can be driven to increase PDT safety by introducing conservative guardbands on the necrosis thresholds of the OAR. Comparing our linear program to the most used algorithm in iPDT planning research, the Cimmino feasibility algorithm, we have shown that our program achieves a damage reduction at the same tumor coverage rate with all photosensitizers considered, which shows how robust the LP is. The reduction ratios varied from 12% – 35% on grey matter and 14% – 54% on white matter (Table 7).
In the second part of this manuscript, we have leveraged from the robustness of this LP and built on top of it an iterative optimization framework that searches and selects from a big space of possible source positions in order to further minimize the damage with possibly less number of sources. The framework introduced returns a set of plans with different qualities and number of sources that a clinician can choose from based on, among other factors, the treatment feasibility and the clinical equipment available. Finally, we have investigated the trade-off between the number of sources and the quality of the plan in terms of tumor coverage and damage to OAR.
7. Conclusion
In this work, we have proposed a fast and effective linear program to be used in iPDT planning for source power allocation. Results on virtual brain tumor models have shown a significant reduction in healthy tissue damage for all photosensitizers and wavelengths tested versus existing power allocation techniques. With the ALCIPc photosensitizer that usually resulted in the best treatment plans due to its high tumor selectivity, our linear program method outperforms the prior Cimmino algorithm by reducing the damage to the white and grey matters by 29% and 31% respectively, while requiring comparable runtime. We have also shown that intuitive changes to the cost function thresholds and importance weights provide an effective way to trade-off between leaving some tumor volume under-treated versus damaging more healthy tissue, which is a useful feature for plan exploration. Additionally, we have proposed an optimization loop that would provide clinicians with a set of high-quality treatment plans with refined source locations. With these plans, a tumor could be fully destroyed with minimal damage to the healthy tissues by ALCIPc mediated PDT with theoretically less than 10 minutes of optical irradiation.
Funding
Ontario Centres of Excellence (OCE) (OCE-24491), Natural Sciences and Engineering Research Council (NSERC) (CRDPJ 490784-15), IBM (OCE-24491), Theralase Technologies Inc. (OCE-24491), and Southern Ontario Smart Computing Innovation Platform (SOSCIP) (OCE-24491).