Performance optimization of real-time neuronavigation

[rmatsuda]

The real-time neuronavigation feature of InVesalius currently works based on multi-threading using Queues, Events, and Jobs (look at #242). Optimization (utilizing the least memory, minimizing its CPU time, and offering high speed) is needed to improve neuronavigation performance.

Deliverables:
Characterize step-by-step code execution time for neuronavigation
Improve threads' sleep times
Investigate: WX GUI becomes slower when navigation is on
Optimize 3D rendering/updating scene

[rmatsuda]

Some examples that can be improved:

• CenterOfMass function
• some self.Refresh can be removed or replaced by self.interactor.Render() (inside the viewer_volume neuronavigation loop)

[Kaushik-Iyer]

After looking at #242 some of the python files have unused imports, will removing them help in this issue?

[jsxs0]

@rmatsuda (@paulojamorim)

CenterOfMass function

I think we can use NumPy to perform the calculations in bulk rather than in a loop for the following (this could significantly improve performance):

    n = surface.GetNumberOfPoints()
    for i in range(n):
        point = surface.GetPoint(i)
        barycenter[0] += point[0]
        barycenter[1] += point[1]
        barycenter[2] += point[2]
    barycenter[0] /= n
    barycenter[1] /= n
    barycenter[2] /= n

    return barycenter

[rmatsuda]

Thanks for the comments! @Kaushik-Iyer removing the imports will not significantly improve the navigation performance. @jsxs0 The numpy could be a good aproach for the center of mass function. Also the vtkCenterOfMass can do the job. I'm not sure which is the best (in terms of performance).

[vhosouza]

@rmatsuda is this issue still valid or the latest updates you made to optimize the performance solve it? Shall we close or investigate further?

[rmatsuda]

The pull request #540 and #600 improved significantly the neuronavigation performance. I will close, thanks!