Is it possible to compute 2D Hessian value for each slice in a 3D medical array?

Beginner Questions
, ,
1

Hi, I am trying to compute 2-dimention hessian value for each slice in a 3D array. I succeed computing 3D hessian value for a 3D image or 2D hessian value for a single slice by using itk.MultiScaleHessianBasedMeasureImageFilter. But is it possible to perform Hessian Augmentation simultaneously for all the slices in a 3D array? I know I could achieve it by writing a for-loop, which is, however, very time-consuming. Thank you for your replies.

Here is how I compute 2D hessian value

def hessian2D(array2d):
    print("*****Hessian-vessel startsļ¼*****")
    start_time = time()
    dimention = len(array2d.shape)
    img_itk = itk.GetImageFromArray(array2d)

    # 1
    sigma_minimum = 0.2
    sigma_maximum = 1  # 3.
    number_of_sigma_steps = 8
    ImageType = type(img_itk)
    Dimension = img_itk.GetImageDimension()
    HessianPixelType = itk.SymmetricSecondRankTensor[itk.D, Dimension]
    HessianImageType = itk.Image[HessianPixelType, Dimension]
    objectness_filter = itk.HessianToObjectnessMeasureImageFilter[HessianImageType, ImageType].New()
    objectness_filter.SetBrightObject(True)
    objectness_filter.SetScaleObjectnessMeasure(True)
    objectness_filter.SetAlpha(0.5)
    objectness_filter.SetBeta(1.0)
    objectness_filter.SetGamma(5.0)
    multi_scale_filter = itk.MultiScaleHessianBasedMeasureImageFilter[ImageType, HessianImageType, ImageType].New()
    multi_scale_filter.SetInput(img_itk)
    multi_scale_filter.SetHessianToMeasureFilter(objectness_filter)
    multi_scale_filter.SetSigmaStepMethodToLogarithmic()
    multi_scale_filter.SetSigmaMinimum(sigma_minimum)
    multi_scale_filter.SetSigmaMaximum(sigma_maximum)
    multi_scale_filter.SetNumberOfSigmaSteps(number_of_sigma_steps)
    output = multi_scale_filter.GetOutput()
    output_array = itk.GetArrayFromImage(output)
    end_time = time()
    print("Hessian-vessel finished, duration:{}s".format(end_time - start_time))
    return output_array
2

Hello @Fivethousand,

Possibly look at the SliceBySliceFilter. Iā€™m not sure it is available in ITKā€™s Python wrapping but likely. The MultiScaleHessianBasedMeasureImageFilter is not available in SimpleITK.

Finally you can always use Python multiprocessing to work on multiple slices in parallel.

1 Like
3

Thank you for your reply, I just notice that most of time is used in C compilation when initializing the

MultiScaleHessianBasedMeasureImageFilter . And the time of iterating each slice could be totally ignored.