Geometry

Geometry#

MITK images carry a spatial geometry (spacing, origin, direction cosines) and an optional time geometry that maps time steps to time points. The bindings expose the geometry both via convenience accessors on mitk.Image and via the full geometry classes.

Accessors on mitk.Image#

The common case is to read and write spacing, origin, and direction for time step 0:

print(img.spacing)            # (sx, sy, sz)
print(img.origin)             # (ox, oy, oz)
print(img.direction)          # 3x3 direction cosine matrix (NumPy array)
print(img.direction_cosines)  # same values, flat 9-element tuple (row-major)

img.spacing = (0.8, 0.8, 1.5)
img.origin  = (0.0, 0.0, -100.0)

For 4D images, per-time-step accessors are also available:

img.get_spacing(time_step=0)
img.set_spacing((0.8, 0.8, 1.5), time_step=0)
img.get_origin(time_step=1)
img.get_direction(time_step=0)
img.set_direction(np.eye(3), time_step=0)

Full geometry objects#

For lower-level access, the full geometry classes are exposed:

geom = img.get_geometry(time_step=0)
print(geom)  # mitk.BaseGeometry subclass

tg = img.time_geometry
print(tg.count_time_steps())
print(tg.get_min_time_point(), tg.get_max_time_point())

Class

Purpose

mitk.BaseGeometry

Common interface for spatial geometries.

mitk.Geometry3D

Concrete 3D spatial geometry.

mitk.PlaneGeometry

Geometry of a 2D plane embedded in 3D.

mitk.SlicedGeometry3D

A stack of plane geometries.

mitk.TimeGeometry

Abstract time geometry.

mitk.ArbitraryTimeGeometry

Per-time-step user-defined time points.

mitk.ProportionalTimeGeometry

Evenly spaced time points.

Time-step / time-point conversion#

TimeGeometry is the bridge between integer time steps (array indices) and floating-point time points (the physical time the user sees):

tg = img.time_geometry
tp = tg.time_step_to_time_point(2)
ts = tg.time_point_to_time_step(tp)
assert tg.is_valid_time_step(ts)
assert tg.is_valid_time_point(tp)

bounds = tg.get_time_bounds()        # (min_tp, max_tp)
geom_t1 = tg.get_geometry_for_time_step(1)
geom_t  = tg.get_geometry_for_time_point(tp)

For static (3D) images, the time geometry has a single time step covering the full time range; the conversions still work and return that one step.

NumPy <-> MITK conventions#

MITK stores spacing, origin, and direction in physical (x, y, z) order. NumPy stores arrays in C (z, y, x) order. The bindings preserve this distinction: img.shape is in NumPy order, but img.spacing, img.origin, and img.direction are in physical order.