Struct Preprocessed 

struct Preprocessed {
    data: Vec<f32>,
    dims: [usize; 3],
    aligned_dims: [usize; 3],
    origin: [isize; 3],
    valid: [(usize, usize); 3],
    perm: [usize; 3],
    flip: [bool; 3],
    source_dims: [usize; 3],
    res: f64,
    out_dims: (usize, usize, usize),
    resampled: bool,
}

The network input plus everything needed to map predictions back onto the caller’s grid.

Fields

data: Vec<f32>

Normalised network input in RAS orientation, C-order [d0, d1, d2]: the padded volume, or the head’s box in it.

dims: [usize; 3]

aligned_dims: [usize; 3]

Shape of the RAS-aligned volume before cropping and padding.

origin: [isize; 3]

Where the network grid’s first voxel sits in the RAS-aligned volume, per axis (negative where the grid starts in padding).

valid: [(usize, usize); 3]

The network-grid range that holds aligned data, [start, end) per axis: inside the padding, and inside any crop.

perm: [usize; 3]

perm[i] is the pre-alignment axis that became RAS axis i.

flip: [bool; 3]

Whether RAS axis i was flipped during alignment.

source_dims: [usize; 3]

Shape of the volume the alignment consumed (post-resampling, pre-alignment).

res: f64

Resolution the network saw, mm.

out_dims: (usize, usize, usize)

The caller’s grid, for mapping the result back.

resampled: bool

Whether a resample to 1 mm happened (the label map is returned on the caller’s grid either way).