Benoît Ozell

Purpose: The purpose of this study was to develop and validate RetinaVR, an affordable, portable, and fully immersive virtual reality (VR) simulator for vitreoretinal surgery training. Methods: We built RetinaVR as a standalone app on the Meta Quest 2 VR headset. It simulates core vitrectomy, peripheral shaving, membrane peeling, and endolaser application. In a validation study (n = 20 novices and experts), we measured: efficiency, safety, and module-specific performance. We first explored unadjusted performance differences through an effect size analysis. Then, a linear mixed-effects model was used to isolate the impact of age, sex, expertise, and experimental run on performance. Results: Experts were significantly safer in membrane peeling but not when controlling for other factors. Experts were significantly better in core vitrectomy, even when controlling for other factors (P = 0.014). Heatmap analysis of endolaser applications showed more consistent retinopexy among experts. Age had no impact on performance, but male subjects were faster in peripheral shaving (P = 0.036) and membrane peeling (P = 0.004). A learning curve was demonstrated with improving efficiency at each experimental run for all modules. Repetition also led to improved safety during membrane peeling (P = 0.003), and better task-specific performance during core vitrectomy (P = 0.038), peripheral shaving (P = 0.011), and endolaser application (P = 0.043). User experience was favorable to excellent in all spheres. Conclusions: RetinaVR demonstrates potential as an affordable, portable training tool for vitreoretinal surgery. Its construct validity is established, showing varying performance in a way that correlates with experimental runs, age, sex, and level of expertise. Translational Relevance: Fully immersive VR technology could revolutionize surgical training, making it more accessible, especially in developing nations.

Simulators for virtual surgery training need to perform complex calculations very quickly to provide realistic haptic and visual interactions with a user. The complexity is further increased by the addition of cuts to virtual organs, such as would be needed for performing tumor resection. A common method for achieving large performance improvements is to make use of the graphics hardware (GPU) available on most general-use computers. Programming GPUs requires data structures that are more rigid than on conventional processors (CPU), making that data more difficult to update. We propose a new method for structuring graph data, which is commonly used for physically based simulation of soft tissue during surgery, and deformable objects in general. Our method aligns all nodes of the graph in memory, independently from the number of edges they contain, allowing for local modifications that do not affect the rest of the structure. Our method also groups memory transfers so as to avoid updating the entire graph every time a small cut is introduced in a simulated organ. We implemented our data structure as part of a simulator based on a meshless method. Our tests show that the new GPU implementation, making use of the new graph structure, achieves a 10 times improvement in computation times compared to the previous CPU implementation. The grouping of data transfers into batches allows for a 80–90% reduction in the amount of data transferred for each graph update, but accounts only for a small improvement in performance. The data structure itself is simple to implement and allows simulating increasingly complex models that can be cut at interactive rates.

One of the main issues when simulating the filling of a mould cavity is not so much solving the equations governing the flow, but rather to make sure that the calculation is accurate, although the shape of the geometrical domain changes constantly. In order to improve the accuracy of the numerical approximation, it is interesting to refine locally the mesh in the neighbourhood of the flow front and project on the new mesh the different scalar fields involved in the calculation. Very often a significant inaccuracy may result from these operations. The numerical error is cumulative and relatively small errors at each time step can lead to significant discrepancies at the end of the simulation. For all these reasons, errors on the conservation of the fluid mass represent a major challenge in injection moulding simulations. In this paper, a new approach based on dual kriging interpolation is proposed to project the flow front from one mesh to another, while conserving automatically the fluid mass in the cavity. The advantage of this methodology dwells in its generality, as it is valid both in the two and three-dimensional cases. It solves a major problem that occurs commonly in the numerical approximation of moving boundary problems, namely the conservation of the fluid mass. It consists of following precisely the displacement of the flow front by updating at each time t a saturation function f(x, t) defined at all points x in the cavity. At any given time, this function is one in the saturated domain, zero in the empty zone and is bounded between zero and one to reflect partial saturation in the vicinity of the flow front. An interpolation of the saturation function based on dual kriging provides an implicit equation of the moving front. From this equation the moving front can then be interpolated on the new mesh, so as to conserve exactly the fluid mass in the mould. After describing the main equations of dual kriging, the article presents the algorithm devised to conserve the fluid mass at each calculation step. Numerical experiments are carried out to validate these concepts for three-dimensional fluid fronts. The conservation of fluid mass is verified for multiple or merging flow fronts, obstacles, ribbed connections and parts of complex geometry.

Soft body deformation models are commonly used in surgery simulations. However, cutting those models can have a severe impact on computation times and affects the interactivity of the simulation. We propose a novel method for modeling topology and introducing cuts in a meshless soft body simulated on a background grid, as well a way to progressively update the visual aspect of the object by adding a small number of triangles to the surface mesh to cover the cut area. We determine that the accuracy of the deformation is preserved after cutting by comparing our method to a finite element method. Tests show that this new method achieves interactive simulation rates with more than 10,000 elements while cutting the model and reconstructing the mesh. Our separation of the visual and physical aspects of the simulation allows for more flexibility when tuning the performance of the simulation. Topology modifications have little impact on computation times for either physical or visual changes.

In proton therapy, Monte Carlo simulations are desirable to accurately predict the delivered dose. This paper introduces and benchmarks pGPUMCD , a GPU-based Monte Carlo code implementing the physical processes required for proton therapy applications. In pGPUMCD , the proton transport is carried out in a voxelized geometry with a class II condensed history scheme. For this purpose, the equivalent restricted stopping power formalism ( L _eq formalism), the Fermi–Eyges scattering theory and the discrete electromagnetic/nuclear interactions were considered. pGPUMCD was compared to Geant4 in a validation study where the physical processes were validated one after the other. Dose differences between pGPUMCD and Geant4 were smaller than 1% in the Bragg peak region and up to 3% in its distal fall-off. Moreover, a voxelwise dose difference below 1% was observed for 99.5% of calculation positions. The pGPUMCD 80% falloff positions matched with those of Geant4 within 0.1%. The pGPUMCD computation times were inversely proportional to the voxel size, with one million protons transported in less than 0.5 s with mm ³ voxels. pGPUMCD , based on the L _eq formalism variance reduction technique, is therefore an attractive candidate for integration in a clinical treatment planning system.

The equivalent restricted stopping power formalism is introduced for proton mean energy loss calculations under the continuous slowing down approximation. The objective is the acceleration of Monte Carlo dose calculations by allowing larger steps while preserving accuracy. The fractional energy loss per step length ϵ was obtained with a secant method and a Gauss–Kronrod quadrature estimation of the integral equation relating the mean energy loss to the step length. The midpoint rule of the Newton–Cotes formulae was then used to solve this equation, allowing the creation of a lookup table linking ϵ to the equivalent restricted stopping power L _eq , used here as a key physical quantity. The mean energy loss for any step length was simply defined as the product of the step length with L _eq . Proton inelastic collisions with electrons were added to GPUMCD , a GPU-based Monte Carlo dose calculation code. The proton continuous slowing-down was modelled with the L _eq formalism. GPUMCD was compared to Geant4 in a validation study where ionization processes alone were activated and a voxelized geometry was used. The energy straggling was first switched off to validate the L _eq formalism alone. Dose differences between Geant4 and GPUMCD were smaller than 0.31% for the L _eq formalism. The mean error and the standard deviation were below 0.035% and 0.038% respectively. 99.4 to 100% of GPUMCD dose points were consistent with a 0.3% dose tolerance. GPUMCD 80% falloff positions ( R 80 ) matched Geant’s R 80 within 1 μ m. With the energy straggling, dose differences were below 2.7% in the Bragg peak falloff and smaller than 0.83% elsewhere. The R 80 positions matched within 100 μ m. The overall computation times to transport one million protons with GPUMCD were 31–173 ms. Under similar conditions, Geant4 computation times were 1.4–20 h. The L _eq formalism led to an intrinsic efficiency gain factor ranging between 30–630, increasing with the prescribed accuracy of simulations. The L _eq formalism allows larger steps leading to a O ( c o n s t a n t ) algorithmic time complexity. It significantly accelerates Monte Carlo proton transport while preserving accuracy. It therefore constitutes a promising variance reduction technique for computing proton dose distributions in a clinical context.

The aim of this study was to evaluate the potential of bGPUMCD, a Monte Carlo algorithm executed on Graphics Processing Units (GPUs), for fast dose calculations in permanent prostate implant dosimetry. It also aimed to validate a low dose rate brachytherapy source in terms of TG-43 metrics and to use this source to compute dose distributions for permanent prostate implant in very short times.The physics of bGPUMCD was reviewed and extended to include Rayleigh scattering and fluorescence from photoelectric interactions for all materials involved. The radial and anisotropy functions were obtained for the Nucletron SelectSeed in TG-43 conditions. These functions were compared to those found in the MD Anderson Imaging and Radiation Oncology Core brachytherapy source registry which are considered the TG-43 reference values. After appropriate calibration of the source, permanent prostate implant dose distributions were calculated for four patients and compared to an already validated Geant4 algorithm.The radial function calculated from bGPUMCD showed excellent agreement (differences within 1.3%) with TG-43 accepted values. The anisotropy functions at r = 1 cm and r = 4 cm were within 2% of TG-43 values for angles over 17.5°. For permanent prostate implants, Monte Carlo-based dose distributions with a statistical uncertainty of 1% or less for the target volume were obtained in 30 s or less for 1 × 1 × 1 mm3 calculation grids. Dosimetric indices were very similar (within 2.7%) to those obtained with a validated, independent Monte Carlo code (Geant4) performing the calculations for the same cases in a much longer time (tens of minutes to more than a hour).bGPUMCD is a promising code that lets envision the use of Monte Carlo techniques in a clinical environment, with sub-minute execution times on a standard workstation. Future work will explore the use of this code with an inverse planning method to provide a complete Monte Carlo-based planning solution.

The purpose of this study was to evaluate the impact of numerical errors caused by the floating point representation of real numbers in a GPU-based Monte Carlo code used for dose calculation in radiation oncology, and to identify situations where this type of error arises. The program used as a benchmark was bGPUMCD. Three tests were performed on the code, which was divided into three functional components: energy accumulation, particle tracking and physical interactions. First, the impact of single-precision calculations was assessed for each functional component. Second, a GPU-specific compilation option that reduces execution time as well as precision was examined. Third, a specific function used for tracking and potentially more sensitive to precision errors was tested by comparing it to a very high-precision implementation. Numerical errors were found in two components of the program. Because of the energy accumulation process, a few voxels surrounding a radiation source end up with a lower computed dose than they should. The tracking system contained a series of operations that abnormally amplify rounding errors in some situations. This resulted in some rare instances (less than 0.1%) of computed distances that are exceedingly far from what they should have been. Most errors detected had no significant effects on the result of a simulation due to its random nature, either because they cancel each other out or because they only affect a small fraction of particles. The results of this work can be extended to other types of GPU-based programs and be used as guidelines to avoid numerical errors on the GPU computing platform.