GPU-Accelerated Interactive Visualization and Planning of Neurosurgical Interventions
Summary (3 min read)
1 INTRODUCTION
- Planning of interventional procedures from preoperative imaging has led to the development of new effective interventional paradigms.
- After selection of the target point, the neurosurgeon defines a set of regions on the surface of the skull, and the system displays a color-coded map that defines the risk associated with the paths within the region.
- With those approaches, the operator needs to select the target point carefully and precisely since the time required by the algorithms for computing optimized paths is generally long (e.g., [2]).
- The possibility of performing this process in real-time or near real-time allows interactive visualization and planning, which would significantly reduce the planning time and thus allow effective intraoperative re-planning of neurosurgical interventions if needed.
- (ii) Implementing those algorithms onto graphics processing units (GPUs).
1.1 Applicability
- Fig. 1(a) illustrates an example of how the proposed approach can assist the operator to select both an optimal insertion and a target point.
- Such dynamic risk maps can significantly facilitate decision-making for the operator.
- For stereotactic robot-assisted neurosurgical interventions, such as in the NeuroArm and NeuroMate systems, the velocity of the tooltip during insertion can be controlled based on the risk function along the path.
- Each risk value along the pathway is computed by setting the needle tip as current target.
- Indeed, as shown in Fig. 1(b), the high speed of this method allows the operator to see the exact risk from the surrounding tissue when the tool is being inserted, by simply placing the target point along the tip of the tool and recomputing the risk map at each step.
2.1 Definition of Planning Risk Maps
- In the formulation of their problem, selection of safe straight access paths entails minimizing a function that quantifies their associated risks.
- The authors define such a function based on the following criteria: (1) the paths cannot intersect any critical tissue, otherwise, this may result in unacceptable patient injury; (2) the paths should be as distant as possible from the critical tissue; and (3) the path length (i.e. the distance between the entry point and the target) should be as short as possible.
- Inspired by [2], the authors also represent structures segmented from imaging as triangular meshes.
- The authors consider the set of candidate paths as the line segments that extend from the vertices of ms to a pre-defined target point p. Calculation of the risk maps involves two computationally expensive steps.
2.2 Acceleration Spatial Data Structures
- To avoid a large bulk of the computation required to obtain the risk map, without sacrificing precision, the authors use acceleration spatial data structures.
- A BVH is a partition of geometric objects into a hierarchy that can be used to accelerate geometric queries as the ones involved in their problem.
- The traversal stops when a triangle hit by the segment is found or when all the triangles within the intersected nodes have been tested.
- In the case of AABB2, the authors can see that only its child labeled AABB2,1 must be tested for intersection.
- For each path the authors determine which of the children of a node is closer, and which one is farther.
2.3 GPU-based Parallel Implementation
- In their proposed approach, the algorithm generates the risk maps by performing the calculations discussed in Section 2.2 for a large number of potential paths.
- The origin of this is that the computations required for each path are highly variable, and this irregular workload leads to underutilization of GPU resources.
- Different blocks are assigned for execution to different SM in a static manner.
- Then, each of the threads, within the warp, processes one of the paths in the task.
- With the centralized queue, none of the SMs is going to be idle, as long as there are remaining tasks to be processed.
3.1 Computational Performance
- To evaluate the efficiency of the proposed approach (i.e., to what degree it can be used in an interactive fashion), the authors measured the time required for performing the computation of the risk maps for meshes with different resolutions.
- The authors implemented their proposed approach on a GPU using CUDA, and timing was recorded using CUDA time events.
- Fig. 5 reports the measured computational times for the above listed range of sizes of the two meshes (i.e. scalp surface and critical tissue).
- The authors observe that even for high-resolution meshes, their proposed approach can achieve interactive rates, and that it is scalable.
- Indeed, as the mesh resolution increases, the number of paths increases linearly.
3.2 Comparison with CPU-based Baseline Implementation
- The authors compared their proposed GPU-accelerated implementation with a CPU-based baseline implementation.
- The CPU-based baseline implementation computes the risk maps by following the same approach that the authors have previously described.
- No attempt was made to make use of the SIMD capability of the CPU in the baseline implementation.
- Interestingly, the GPU implementation also scales better to the number of computed paths and to the size of the vital tissue mesh than the CPU-based baseline implementation.
- That is, although both the implementations are algorithmically equivalent, the speed-up increases along with the size of the problem.
3.3 Performance Improvement due to the Task Queue
- A speed-up of up to 1.4x (i.e. a 29% reduction in computational time) was measured for a GPU implementation using the centralized task queue with respect to a baseline GPU implementation (see Fig. 6).
- In the baseline GPU implementation the authors configure a grid that holds as many threads as there are paths to process.
- As shown in Fig. 6 the achieved speed-up increases with the size of the vital structure mesh, but is not significantly sensitive to the number of computed paths.
- The explanation for this is that when the size of the mesh is increased the application becomes memory bounded, that is, the computation becomes dominated by the time spent on memory operations.
3.4 Comparison with Voxel-based Approach
- The authors also compared their proposed approach to a fast voxel-based method for computing the risk maps [3] running on the CPU.
- The total risk associated with each path is then computed as the sum of the risks of all the voxels intersected by the surgical instrument.
- Since the input structure of the two methods are fundamentally different, the authors make the following assumption: the vital tissue mesh and the volumetric data are equivalent in terms of size, if the mesh is the iso-surface reconstructed from a segmented 3D image with the same resolution as the volumetric data by the classical marching cubes algorithm.
- The voxel-based approach has the additional disadvantage of growing linearly with the size of the input structure.
- This leads to a quick loss of the interactive rate for the voxel-based approach, as shown in Fig. 7.
3.5 User Study
- The authors conducted a user study in order to quantitatively assess the benefits that interactive computation of risk maps might have on the planning of straight access neurosurgical interventions (Fig. 9).
- For the visually-guided treatment, the subjects were asked to find the best path they could find by selecting an insertion point and freely positioning the target without the aid of risk maps.
- In addition to this, for each tried path, they were presented with information about the length and the proximity to blood vessels.
- For the risk-map guided target, the user was allowed to use guidance from the risk map in order to select a satisfactory target position.
- This computation has to be performed only once for a given region of interest.
3.6 Neurosurgeon Evaluation
- The ultimate goal of their work is to be able to provide assistance to the neurosurgeon for intra-operative planning/re-planning.
- Interactive preoperative and intra-operative target repositioning were identified as highly beneficial features.
- Integration between the planning assistance software, the imaging acquisition mechanisms, and the surgical plan execution tool is the key in the intra-operative scenario.
- Another related and important application that could be benefited by the proposed computational framework is the procedures using the gamma knife, which is used to apply radiation doses over tumors.
Did you find this useful? Give us your feedback
Citations
34 citations
28 citations
21 citations
Cites methods from "GPU-Accelerated Interactive Visuali..."
...Currently the algorithms are written in C++, incorporating appropriate image processing libraries (such as ITK), and optimizing it with multithread implementation [13] and GPU acceleration [14]....
[...]
21 citations
Cites methods from "GPU-Accelerated Interactive Visuali..."
...The hardware/GPU acceleration and/or software-based preprocessing of the data to facilitate interactive selection of trajectory are described in [13,18,19]....
[...]
19 citations
Cites methods from "GPU-Accelerated Interactive Visuali..."
...In [26], the estimated risk associated to accessing path was visualized interactively thanks to GPU optimized methods....
[...]
References
413 citations
334 citations
107 citations
"GPU-Accelerated Interactive Visuali..." refers background in this paper
...Index Terms—GPU Acceleration, Neurosurgical Interventions, Risk Maps, and Straight Access...
[...]
Related Papers (5)
Frequently Asked Questions (15)
Q2. What have the authors stated for future works in "Gpu-accelerated interactive visualization and planning of neurosurgical interventions" ?
In the future, the authors plan to further investigate efficient approaches for procedures on dynamically changing structures, secondary to breathing, cardiac beating, and/or to the interventional procedure. Another future research direction would be to extend the current work for efficient planning and visualization of surgical interventions with non-straight access paths.
Q3. What is the key in the intra-operative scenario?
Integration between the planning assistance software, the imaging acquisition mechanisms, and the surgical plan execution tool is the key in the intra-operative scenario.
Q4. What is the problem of a naive scheme of assigning work to processing units?
With respect to the problem of computing the risk map, a naive scheme of assigning work to processing units in CUDA or OpenCL would correspond to launch the execution of a grid with as many threads as the paths.
Q5. How can the authors safely discard traversals through a node?
Since the closest vertex is closer to their path than the closest point in bounding box AABB2,2, the authors can safely discard traversals through such node.
Q6. What is the disadvantage of the voxel-based approach?
Although the computational time of both the mesh-based and voxel-based approaches is approximately linear to the number of computed paths, the voxel-based approach has the additional disadvantage of growing linearly with the size of the input structure.
Q7. What is the current clinical workflow for using the herein proposed approach?
Their current clinical workflow for using the herein proposed approach entails interactively modifying (with a pointing device such as a mouse) the location of the target point and visualizing the effects on the risk maps.
Q8. What is the common practice of surgeons to use to plan a surgical procedure?
A common practice is manual selection and assessment of the suitability of different entrance positions on the scalp surface of the patient in order to select an appropriate insertion path.
Q9. What is the advantage of using a GPU-accelerated method to plan and visualize neurosurgical interventions?
The possibility of performing thisprocess in real-time or near real-time allows interactive visualization and planning, which would significantly reduce the planning time and thus allow effective intraoperative re-planning of neurosurgical interventions if needed.
Q10. What are the two options for interactive visualization and planning?
Testing each path for intersection against each triangle, and computing its closest distance to each vertex are not affordable options for interactive visualization and planning.
Q11. What is the way to compute a risk map?
An efficient GPU implementation of their approach further enables the computation of risk maps at an interactive rate even for high-resolution meshes.
Q12. What is the way to visualize the path?
When the operator selects a path, the path is usually further analyzed and inspected by looking at imaging planes (MR slices) orthogonal to the insertion path (also known as bird’seye view).
Q13. What are the criteria for a path to be defined?
The authors define such a function based on the following criteria: (1) the paths cannot intersect any critical tissue, otherwise, this may result in unacceptable patient injury; (2) the paths should be as distant as possible from the critical tissue; and (3) the path length (i.e. the distance between the entry point and the target) should be as short as possible.
Q14. What is the problem with the SMs?
the workload assigned to some blocks will need moretime than other blocks, eventually leading to a situation where some of the SMs are idle for a portion of the overall computation time.
Q15. What is the definition of long-running warps?
In their work, long-running warps are those that only finish their execution when all the paths in the global queue have been processed.