The Science

---

Multi-Planar Reconstruction

A volumetric medical image is a 3-D array of voxels. Each voxel has a physical size in millimetres stored in the file header. MPR slices through this array in three orthogonal directions:

\[\text{Axial}(z) = V[z, :, :] \qquad \text{Coronal}(y) = V[:, y, :] \qquad \text{Sagittal}(x) = V[:, :, x]\]

Where \(V\) has shape \((Z, Y, X)\).

Anatomical orientation:

Plane	Slice direction	Standard orientation
Axial	Inferior → Superior	Superior at top
Coronal	Anterior → Posterior	Superior at top
Sagittal	Left → Right	Superior at top

MLenz flips Coronal and Sagittal slices vertically before display so superior is always at the top — matching clinical convention.

Crosshair synchronisation

Dragging the crosshair to \((x, y)\) in the Axial plane causes the Coronal plane to display slice at column \(x\), and Sagittal to display slice at column \(y\). All three planes always show the same physical voxel.

---

Window and Level

Medical images span wider intensity ranges than a display can show. Brain CT covers roughly −1000 HU (air) to +3000 HU (bone). Mapping this linearly compresses soft tissue into ~6% of the display range.

Window/Level selects a sub-range to fill the full display:

\[\text{pixel} = \text{clip}\!\left(\frac{I - (L - W/2)}{W},\ 0,\ 1\right)\]

Where \(I\) is voxel intensity, \(L\) is Level (centre), \(W\) is Window (width).

Standard CT presets (Hounsfield Units):

Tissue	L	W
Brain soft tissue	40	80
Subdural haematoma	75	215
Bone	400	1800
Lung	−600	1500

MLenz normalises all intensities to [0, 1] on load. L and W are expressed as fractions. The default (L=0.5, W=1.0) shows the full range.

---

VTK GPU Volume Rendering

Volume rendering casts a ray from each screen pixel through the volume and integrates contributions along the path.

Transfer functions control how intensity maps to colour and opacity:

• Colour transfer function → intensity to RGB
• Opacity transfer function → intensity to transparency

MLenz uses vtkGPUVolumeRayCastMapper — GPU-accelerated OpenGL ray casting, the same approach used by 3D Slicer and OsiriX.

Transfer function presets

Preset	Colour strategy	Opacity strategy
mri_default	Black → grey → white	Transparent < 0.1, opaque > 0.7
bone	Black → tan → white	Only high intensities (> 0.6) visible
angio	Black → red → orange	Mid-to-high intensities visible
pet	Blue → cyan → yellow → white	Low counts transparent

---

NIfTI format

NIfTI stores a 3-D intensity array plus a 348-byte header containing voxel dimensions, data type, and an orientation matrix (qform/sform) that maps voxel indices to millimetre coordinates in patient space.

MLenz uses SimpleITK's orientation correction to reorient all loaded volumes to a consistent LPS+ orientation before display.

---

DICOM loading

DICOM stores each 2-D slice as a separate file. In the UI, MLenz loads a single .dcm file for quick inspection. The core loader still supports full series loading for scripts and notebooks. Sorting by filename is unreliable, so series loading uses SimpleITK's GetGDCMSeriesFileNames() which sorts by the ImagePositionPatient DICOM tag — the physical position of each slice guaranteed by the scanner — giving correct spatial order.

Physical voxel spacing from PixelSpacing and SliceThickness tags is applied to the viewport aspect ratio so non-isotropic acquisitions (e.g. thick axial slices) display with correct proportions.