SU‐E‐T‐655: Dose Matrix Sampling Heuristic Applied to Improving IMRT Optimization Efficiency

Aim: Most of targeted cancer drugs are developed in the clinical setting as single agent or as add-on of a standard treatment. Indeed, scarce data are available about the interaction effect in terms of synergism or antagonism for combinations of targeted agent or of targeted agents and cytotoxics at different doses. We studied the interaction effects of combinations of BKM120, MEK162 and SN38 (the active metabolite of irinotecan) on CRC cell lines. Materials and methods: A combination drug screening was performed on 7 CRC cell lines, 2 of which previously made resistant to irinotecan. For each cell line the combinations of BKM120 and MEK162, of BKM120 and SN38, and of MEK162 and SN38 were tested. Cell survival was assessed by sulforhodamine B cytotoxic assay in matrices of full-range dose combinations for both drugs. We compiled an R script to perform a synergy analysis according to a modified version of the method proposed by Léhar and al. (Nat Biotechnol 2009;27:659-666). Our version of the Léhar method allows quantification of synergism or antagonism for each dose combination point in dose matrices of two or more drugs, and yields a synthetic interaction parameter, the combination index (CI). A CI of > 0 indicating synergy, a CI of < 0 antagonism, and a CI of 0 additivity. Results: We could detect a pronounced synergism between BKM120 and MEK162, with a mean combination index of 1.82 for the 7 cell lines tested. Drug combinations including chemotherapy showed less positive interactions. A moderate synergistic interaction was detected between SN38 and MEK162, with a mean CI of 1.16. Combinations of SN38 and BKM120 were mainly additive, with a mean CI of 0.17. Interestingly, the distribution of synergistic effect was not homogeneous over the dose matrices. Analysis of matrices for the BKM120/MEK162 combination showed focal areas high synergy, which were quite constantly located in dose regions corresponding to IC10-IC30 for BKM120 and IC20-IC35 for MEK162. On the contrary, analysis of matrices for the BKM/SN38 combination showed focal areas of intense antagonism located in dose regions corresponding to IC15-IC30 for BKM120 and IC5-IC30 for SN38. No clear correlation was detected between, on the one side, the type of pharmacological interaction for all drug combinations tested and, on the other side, the mutational status of KRAS and p53, or sensitivity to irinotecan. Conclusions: Our data show the importance of an extensive preclinical testing of the interaction effects of drug combinations at different doses. We are now performing synergism analysis of three dimension matrices for three-drug combinations. Phosphoproteomic analysis of drug combination effect on this high synergism areas compared to other dose matrix areas is ongoing, with the aim to elucidate the underlying molecular mechanism(s). Citation Format: Diego Tosi, Salima Atis, Caroline Mollevi, Nadia Vie, Pierre Martineau, Celine Gongora. Synergism analysis of dose matrices for combinations of BKM120 (a PI3K inhibitor), MEK162 (a MEK1/2 inhibitor) and chemotherapy in colorectal cancer (CRC) cell lines. [abstract]. In: Proceedings of the AACR Precision Medicine Series: Drug Sensitivity and Resistance: Improving Cancer Therapy; Jun 18-21, 2014; Orlando, FL. Philadelphia (PA): AACR; Clin Cancer Res 2015;21(4 Suppl): Abstract nr A40.

SUMMARY Many interesting inverse problems in geophysics are non-linear and multimodal. Parametrization of these problems leads to an objective function, or measure of agreement between data and model predictions, that has a complex topography with many local minima. Optimization algorithms that rely on local gradients in the objective function or that search the model space locally may become trapped in these local minima. By combining simulated annealing with the downhill simplex method, a hybrid global search algorithm is presented in this paper for non-linear, multimodal, inverse problems. The hybrid algorithm shares the advantages of both local search methods that perform well if the local model is suitable, and global methods that are able to explore efficiently the full model space. The hybrid algorithm also utilizes a larger and more complex memory to store information on the objective function than simulated annealing algorithms. The effectiveness of this new scheme is evaluated in three problems: minimization of the multidimensional Rosenbrock function, non-linear, 1 -D, acoustic waveform inversion, and residual statics. The performance of the hybrid algorithm is compared with simulated annealing and genetic algorithms and is shown to converge more rapidly and to have a higher success rate of locating the global minimum for the cases investigated.

Purpose: Planning roboticradiosurgery for patients with 3 or more metastatic brain lesions is due to the need for a large number of constraints and the requirement of prescribing different doses to separate targets. A two‐step optimization technique was developed to improve the plan quality of such treatments.Method and Materials: First, separate treatment plans were developed for individual targets via setting nominal doses to each target disregarding contributions from other lesions. Then the 3D dose matrix associated with each target was exported and a singular‐value‐decomposition algorithm was applied to balance the background dose changes as well as dose interferences among targets via adjusting relative weightings among the dose matrices. Comparisons were carried out for such approach versus a planning‐for‐all approach implemented on the latest commercial system. Results: Significant improvements were found in the target volume coverage with the new technique as compared with the existing plan‐for‐all techniques due to the capability to select individualized isodose levels for each target. Additionally, improvement was particularly noted for low‐level peripheral normal brain sparing. On average, the volume of the normal brain receiving more than 4‐Gy was reduced by 13% with the new technique for the studied cases. The improvement was especially notable when a high number (n>5) of targets are considered. Conclusion: A two‐step optimization technique has been demonstrated to improve treatment planning quality as well as the flexibility of dose prescription for multi‐target treatment planning for roboticradiosurgery.

The purpose is to calculate the composite 3D biological effective dose (BED) distribution in healthy liver, when multiple lesions are treated concurrently with different hypo-fractionated schemes and stereotactic body radiation therapy, and to investigate the potential of biological based plan optimization. Two patients, each having two tumors that were treated sequentially with different treatment plans, were selected. The treatment information of both treatment plans of the patients was used and their dose matrices were exported to an in-house MATLAB software, which was used to calculate the composite BED distribution. The composite BED distributions were used to determine if the healthy liver received BED beyond tolerance. When the dose to the minimum critical volume was less than tolerance, an optimization code was used to derive the scaling factors (ScF) that should be applied to the dose matrix of each plan until the minimum critical volume of healthy liver reaches a BED close to tolerance. It was shown that for each patient, there is a margin for dose escalation regarding the doses to the individual targets. More specifically, the ScFs of the doses range between 5.6 and 99 in the first patient, whereas for the second patient, the ScFs of the optimal doses range between 12.7 and 35.6. The present study indicates that there is a significant margin for dose escalation without increasing the radiation toxicity to the healthy liver. Also, the calculation of the composite BED distribution can provide additional information that may lead to a better assessment of the liver's tolerance to different fractionation schemes and prescribed doses as well as more clinically relevant treatment plan optimization.

Purpose: To develop a tool that enables us to process the dose matrix generated by the same or different radiation treatment planning system. Method and Materials: A Matlab tool was developed in order to post‐process the dose matrix generated from different treatment planning systems. Currently, the tool can import the dose matrix from Pinnacle3, Tomotherapy, Nomos Corvus and gamma knife planning computers. Tumor contours can also be imported. Using the tool one can calculate region‐of‐interest dose statistics, dose volume histograms (DVH), 2D dose maps, 3D dose maps, effective uniform dose (EUD), biological uniform dose (BUD), and dose profiles. The software tool can produce a comprehensive report that includes all the calculation. The software has been used to facilitate a comparison study between Tomotherapy and Gamma Knife for radiosurgical cases. The conformality index, defined as the ratio of the prescribed dose divided by target volume, was used to evaluate the treatment plans. Maximum dose, minimum dose, means dose, DVH and treatment time were also parameters used in the comparison. Results: Five gamma knife patients have been selected as candidates for the dose comparison between tomotherapy and gamma knife. All five patients had single brain lesions. Results show that tomotherapy can achieve the same DVH with higher dose conformality as compared to GK. However, the treatment volume of the lower dose lines (<10 Gy) is higher than gamma knife. Conclusion: A Matlab tool has been developed for post‐processing dose matrices generated by different treatment planning systems. The tool has been applied on comparison of the radio surgery cases between tomotherapy and gamma knife. The comparison results show that tomotherapy can deliver radio surgical precision and achieve the same results as gamma knife considering conformality index, maximum dose, minimum dose and mean dose in the tumor.

Study on the stability of lung cancer radiotherapy dosiomic features under respiratory motion.

In this paper, we propose a fast rigid-registration method of inferior limb X-ray films (two-dimensional Computed Radiography (CR) images) and three-dimensional Computed Tomography (CT) images for Total Knee Arthroplasty (TKA) surgery planning. The position of the each bone, such as femur and tibia (shin bone), in X-ray film and 3D CT images is slightly different, and we must pay attention how to use the two different images, since X-ray film image is captured in the standing position, and 3D CT is captured in decubitus (face up) position, respectively. Though the conventional registration mainly uses cross-correlation function between two images,and utilizes optimization techniques, it takes enormous calculation time and it is difficult to use it in interactive operations. In order to solve these problems, we calculate the center line (bone axis) of femur and tibia (shin bone) automatically, and we use them as initial positions for the registration. We evaluate our registration method by using three patient's image data, and we compare our proposed method and a conventional registration, which uses down-hill simplex algorithm. The down-hill simplex method is an optimization algorithm that requires only function evaluations, and doesn't need the calculation of derivatives. Our registration method is more effective than the downhill simplex method in computational time and the stable convergence. We have developed the implant simulation system on a personal computer, in order to support the surgeon in a preoperative planning of TKA. Our registration method is implemented in the simulation system, and user can manipulate 2D/3D translucent templates of implant components on X-ray film and 3D CT images.

In PCB routing and package routing, routes must satisfy various constraints and specifications. In recent years, an any-angle routing using a quasi-Newton method has been proposed to obtain a routing pattern satisfying the constraints and specifications. However, computational time of the quasi-Newton method is long because many computations are iterated for the evaluation of the objective function and its gradient. In this paper, we propose an acceleration method for the any-angle routing using a quasi-Newton method by GPGPU. Moreover, we also propose an improved method by omitting the redundant calculation of the evaluation of the objective function. Experiments show that the proposed method is effective by the comparison on the quality of obtained routes and the computational time.

Optimized non-coplanar beam ensembles may yield substantial improvements of radiation therapy treatment plan quality. However, beam angle selection (BAS) is a nontrivial process and automated beam ensemble optimization is not supported by most commercial treatment planning systems, in part due to the lack of efficient algorithms. We focus on iterative BAS, where a beam ensemble is sequentially constructed: the basis for BAS during iteration n is a treatment plan containing n-1 fixed beams. To select the nth beam, each remaining candidate beam is added one-by-one to the current beam ensemble and the fluence map optimization (FMO) problem is solved. The beam yielding the lowest objective function value is selected. In this work we investigate two alternative methods relying on objective function surrogates to score candidate beams in order to make iterative BAS computationally more efficient: Conventional iterative BAS is compared to (1) an approach where the FMO is not solved to optimality but stopped after five iterations of a gradient based algorithm and (2) an approach where candidate beams are scored based on a projected gradient of the FMO in the first iteration. In a treatment planning study including one pancreas, one prostate, and one intracranial case, it is observed that both iterative BAS methods using objective function surrogates yield similar plan quality compared to naive iterative BAS with regard to the resulting the objective function values and DVHs. At the same time, the surrogates enable reductions in computation time by a factor of 50-100 making them highly attractive for clinical application.

This article integrates the company operations decisions (i.e. location, production, inventory, distribution, and transportation) and finance decisions (i.e. cash, accounts payable and receivable, debt, securities, payment delays, and discounts) in which the demands and return rate are uncertain, defined by a set of scenarios. The cash flow and budgeting model will be coupled with supply chain network design using a mixed integer linear programming formulation. The article evaluates two financial criteria, that is, the change in equity and the profit as objective functions. The results indicate that objective functions are partially interdependent, that is, they conflict in certain parts. This fact illustrates the inadequacy of treating process operations and finances in isolated environments and pursuing objective myopic performance indicators such as profit or cost. Due to the importance of the supply chain network design problem, a multi-objective robust optimization with the max–min version is extended to cope with the uncertainty. A solution approach integrating Benders’ decomposition method with the scenario relaxation algorithm is also proposed in this research. The improved algorithm has been applied to solve a number of numerical experiments. All results illustrate significant improvement in computation time of the improved algorithm over existing approaches. For a problem, the proposed algorithm shows a significant reduction in computational time compared with the Benders’ decomposition and scenario relaxation that shows the efficiency of the proposed solution method.

Optimization of FRP composites against impact induced failure using island model parallel genetic algorithm

Database sequencing applications including sequence comparison, searching, and analysis are considered among the most computation power and time consumers. Heuristic algorithms suffer from sensitivity while traditional sequencing methods, require searching the whole database to find the most matched sequences, which requires high computation power and time. This paper introduces a dynamic programming technique based-on a measure of similarity between two sequential objects in the database using two components, namely frequency and mean. Additionally, database sequences that have the lowest scores in the comparison process were excluded such that the similarity algorithm between a query sequence and other database sequences is applied to meaningful parts of the database. The proposed technique was implemented and validated using a heterogeneous HW/SW FPGA-based embedded system platform. The implementation was partitioned into (1) hardware part (running on logic gates of FPGA) and (2) software part (running on ARM processor of FPGA). The validation study showed a significant reduction in computation time by accelerating the database sequencing processes by 60% comparing to traditional known methods.

This paper analyses different objective functions for the parameter identification of parallel mechanisms and studies the influence in the position and orientation errors to improve their accuracy. A new objective function considering deviation terms is presented. This function is compared with other widely used functions and the advantages and disadvantages of each function are presented. The geometric parameter identification is performed by external calibration by means of the direct kinematic model. First, the objective functions are defined considering error position, error orientation and deviations in measurement. These functions compare the measured and calculated moving platform coordinates in order to obtain the identified model parameters that minimize this difference. The measured coordinates are obtained by measuring three sphere centres, fixed to the moving platform of a parallel mechanism, and the computed coordinates are given by the kinematic model. Second, the model is solved by the Levenberg–Marquardt algorithm for a number of identification positions. Finally, the calibration is verified in test positions. The results obtained show that the consideration of the deviations in measurement in the objective function with respect to classical approaches allows us to better identify those kinematic parameters corresponding with passive joints that cannot be measured. These findings confirm that a suitable objective function can improve the mechanism accuracy by more than one order of magnitude in both position and orientation errors.

Increasing demand for containerization compels container terminals to improve their performance. Uncoordinated scheduling of operations is one of the main factors accounting for poor performance at automated container terminals (ACTs). To increase land utilization efficiency and lower operational times, a new storage system called the split-platform automated storage/retrieval system (SP-AS/RS) has been introduced for temporary storage of containers. This paper describes a multi-objective mixed-integer programming (MIP) model that is based on a combination of multiple interacting sub-tasks. It is aimed at optimizing the integrated scheduling of handling and storage operations in ACTs. The MIP model objective function is to minimize delays in the loading/unloading tasks of the cranes and the travel time of vehicles and platforms in the SP-AS/RS. At the same time, a simulated annealing algorithm (SAA) that provides near-optimal solutions for the problem in a reasonable computation time is appraised. The results of this study show that the objective function of the MIP model is, on average, 58 % lower than that of the non-integrated scheduling method. On the other hand, the best objective function values obtained by the SAA indicate only a 3.7 % disadvantage in comparison with optimal values determined by the MIP model, demonstrating that the SAA is able to provide near-optimal solutions for the integrated scheduling of handling and storage operations.

A stochastic algorithm is presented for finding the global optimum of a function of n variables subject to general constraints. The algorithm is intended for moderate values of n , but it can accommodate objective and constraint functions that are discontinuous and can take advantage of parallel processors. The performance of this algorithm is compared to that of the Nelder-Mead Simplex algorithm and a Simulated Annealing algorithm on a variety of nonlinear functions. In addition, one-, two-, four-, and eight-processor versions of the algorithm are compared using 64 of the nonlinear problems with constraints collected by Hock and Schittkowski. In general, the algorithm is more robust than the Simplex algorithm, but computationally more expensive. The algorithm appears to be as robust as the Simulated Annealing algorithm, but computationally cheaper. Issues discussed include algorithm speed and robustness, applicability to both computer and mathematical models, and parallel efficiency.