Parallel radiation dose computations with GENOCOP III on GPUs

Moreno, Juan José Hurtado; Miroforidis, Janusz; Filatovas, Ernestas; Kaliszewski, Ignacy; Garzón, Eduardo Garzón

doi:10.1007/s11227-020-03254-6

Cited by 4 publications

(3 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The first level of the iterative process is managed by the evolutionary optimization algorithm, which controls the population of candidate solutions by applying variation operators such as mutation, crossover, and selection [12]. At the second level, the new generation of solutions is evaluated using the custom-coded GD [10]. This evaluation takes patient data, deposition matrix, beamlet geometry, ROI segmentation, and parameters φ as inputs and returns optimal fluence x ⋆ (φ).…”

Section: Evolutinary Methods To Adjust the Geud Parametersmentioning

confidence: 99%

“…As seen in Section 2.1 and in [10], the GD method to calculate x ⋆ (φ) consists of an iterative process that computes matrix vector products with the large and sparse deposition matrix, D and its transpose D T . This computation is successively performed for each individual evaluated in the population and dominates the huge computational cost of PersEUD due to the large dimensions of D. Therefore, to improve its performance, our focus is on the products of the deposition matrix that characterize every patient, bearing in mind that memory management has a strong impact on the performance of these operations.…”

Section: Automated Geud Tuning On Multicorementioning

confidence: 99%

“…In practice, the values of these parameters are initially fixed from the literature and then adjusted by trial and error where the GD method is involved to obtain a clinically feasible plan. It is relevant to emphasize that the computational cost of the GD method customized to the gEUD model is dominated by the operations with the deposition matrix D and its transpose D T at every step of the GD algorithm, as described in [10].…”

Section: Parameters To Adjust In the Geud Modelmentioning

confidence: 99%

See 2 more Smart Citations

Leveraging Multicore Servers for Enhanced IMRT Radiotherapy Planning

Riado,

Martín,

Redondo

et al. 2023

Preprint

View full text Add to dashboard Cite

This study enhances the computational efficiency of Radiotherapy Plans (RP) utilized in Intensity Modulated Radiotherapy (IMRT). IMRT is a technique that employs radiation beams with varying angles and intensities to deliver radiation to cancerous tissues while safeguarding healthy organs. The planning methods reliant on the gEUD metric yield radiation plans with exceptional PTV (Planning Target Volume) coverage.Nevertheless, computing these plans is a resource-intensive task that entails adjusting numerous parameters and conducting multiple model evaluations. To address this, we have introduced a novel approach that automates the adjustment of gEUD parameters. This is achieved by combining the EUD model, solved through a gradient descent algorithm, with an evolutionary optimization method that explores the EUD parameter space.Given the high computational demands of this approach, integrating it into clinical settings poses a challenge. Our goal is to tackle this challenge by introducing parallelization and batching strategies that leverage the capabilities of multicore servers, aiming to significantly accelerate the optimization process.To evaluate our proposal, we conducted extensive benchmarking on three distinct multicore platforms with varying micro-architectures, assessed across different batch sizes and threads configurations. Our testing dataset consisted of three Head and Neck (H\&N) patients who were treated using IMRT with nine beams.The results showcase that our approach provides substantial computational speed improvements while consistently generating high-quality RT (Radiation Therapy) plans that conform to clinical constraints.

show abstract

Section: Evolutinary Methods To Adjust the Geud Parametersmentioning

confidence: 99%

Section: Automated Geud Tuning On Multicorementioning

confidence: 99%

Section: Parameters To Adjust In the Geud Modelmentioning

confidence: 99%

See 1 more Smart Citation