Cinematic heart

References

• volume

• plot

• Cinematic rendering

heart.mhd heart.zraw

A cardiac CT as a participating medium rather than a ray march: light scatters inside the tissue, the denser boundaries shade as surfaces, and what lights them is the environment.

import k3d
import numpy as np
import SimpleITK as sitk

im_sitk = sitk.ReadImage('heart.mhd')
img = sitk.GetArrayFromImage(im_sitk)
size = np.array(im_sitk.GetSize()) * np.array(im_sitk.GetSpacing())

# OrRd_r with its colour channels lifted above 1: the transfer function is what the
# medium emits and reflects, so values over unity read as brighter tissue
color_map = (np.array(k3d.colormaps.matplotlib_color_maps.OrRd_r).reshape(-1, 4)
             * np.array([1, 1.25, 1.25, 1.25])).astype(np.float32)

plt_volume = k3d.volume(img.astype(np.float16),
                        alpha_coef=250,
                        samples=256,
                        light_scale=2.25,
                        color_range=[300, 900],
                        color_map=color_map)

plt_volume.transform.bounds = [-size[0] / 2, size[0] / 2,
                               -size[1] / 2, size[1] / 2,
                               -size[2] / 2, size[2] / 2]

plot = k3d.plot(renderer='cinematic',
                environment='neutral',
                background_color=0x2A2C30,
                grid_visible=False,
                camera_auto_fit=False,
                colorbar_object_id=0,
                cinematic_samples=256,
                cinematic_bounces=5,
                cinematic_denoise=2.0,
                cinematic_bokeh_size=10.0,
                cinematic_focus_distance=185.2,
                lighting=1.5)
plot += plt_volume
plot.display()

# aimed at the centre of mass of what the colour range keeps, not at the centre of the
# data box - the tissue is nowhere near it
plot.camera = [-3.3, -221.6, 91.6, -3.3, 7.1, 17.3, 0, 0, 1]

The four knobs worth turning first. alpha_coef is how much matter a ray meets, and it sets the cost as well as the density - lower it and light reaches deeper. light_scale is the medium’s own exposure, separate from plot.lighting, because a dense volume needs more light than the geometry around it. color_range decides what counts as tissue at all. And cinematic_bounces is what carries light into the interior: at 1 the inside goes black however bright the environment.

The lens is the other half of the look. The tracer samples a real aperture, so depth of field costs no extra samples - though a defocused frame wants more of them, because it averages over a wider set of paths. cinematic_focus_distance is set here rather than left at 0: zero focuses on whatever the camera is pointed at, which is the centre of the data box, and that sits behind the surface anyone is actually looking at. The value here is about three quarters of the camera’s own distance, which puts the plane on the front of the tissue and leaves the ribs behind it soft.