Compared With NIfTI

1. Problem: NIfTI’s Core Challenges In Clinical Workflows

NIfTI is the de facto format in neuroimaging research, enabling sharing and algorithm development. However, when used in rigorous clinical workflows, three issues become prominent: inconsistent spatial conventions, heavy metadata loss, and limited compression efficiency.

These issues compound across the lifecycle from acquisition (DICOM) → research (NIfTI) → back to clinical systems (e.g., navigation/archiving), creating obstacles for consistency and traceability.

1.1. Spatial Inconsistency Across Libraries

Reading the same NIfTI file using different libraries can yield inconsistent origin and orientation (RAS+ vs LPS+ practices), leading to downstream registration and analysis risks.

import nibabel as nib
import SimpleITK as sitk
import numpy as np

nifti_file = "dicube-testdata/nifti/CT_Philips.nii.gz"
nib_image = nib.load(nifti_file)
sitk_image = sitk.ReadImage(nifti_file)

print(f"Nibabel Origin: {nib_image.affine[:3, 3]} (RAS+ oriented)")
print(f"SimpleITK Origin: {sitk_image.GetOrigin()} (LPS+ oriented)")

x_nib = nib_image.affine[:3, 0]; x_nib /= np.linalg.norm(x_nib)
print(f"Nibabel X Orientation: {x_nib}")
print(f"SimpleITK X Orientation: {np.array(sitk_image.GetDirection())[:3]}")

y_nib = nib_image.affine[:3, 1]; y_nib /= np.linalg.norm(y_nib)
print(f"Nibabel Y Orientation: {y_nib}")
print(f"SimpleITK Y Orientation: {np.array(sitk_image.GetDirection())[3:6]}")

z_nib = nib_image.affine[:3, 2]; z_nib /= np.linalg.norm(z_nib)
print(f"Nibabel Z Orientation: {z_nib}")
print(f"SimpleITK Z Orientation: {np.array(sitk_image.GetDirection())[6:9]}")

Nibabel Origin: [ -82.32080078 -134.36405945 -153.72033691] (RAS+ oriented)
SimpleITK Origin: (86.97265625, 229.64453125, 192.01113891601562) (LPS+ oriented)
Nibabel X Orientation: [1. 0. 0.]
SimpleITK X Orientation: [-1.  0.  0.]
Nibabel Y Orientation: [0. 1. 0.]
SimpleITK Y Orientation: [ 0. -1.  0.]
Nibabel Z Orientation: [0. 0. 1.]
SimpleITK Z Orientation: [0. 0. 1.]

1.2. Coordinate Systems: LPS+ vs RAS+

• LPS+ (DICOM/clinical): X→Left, Y→Posterior, Z→Superior
• RAS+ (neuro/research): X→Right, Y→Anterior, Z→Superior

They differ on X and Y directions, explaining the common sign flips across libraries.

1.3. Dual Affines: qform vs sform

NIfTI stores two affine transforms: qform (scanner space) and sform (aligned/standard space). Libraries may prefer one over the other, producing varying origins/orientations for the same file.

qform = nib_image.get_qform()
sform = nib_image.get_sform()
print("qform (scanner space):\n", np.round(qform, 3))
print("\nsform (standard space):\n", np.round(sform, 3))

qform (scanner space):
 [[   0.961    0.       0.     -86.973]
 [   0.       0.964    0.    -229.645]
 [   0.       0.       0.963  192.011]
 [   0.       0.       0.       1.   ]]

sform (standard space):
 [[   0.961    0.       0.     -82.321]
 [   0.       0.964    0.    -134.364]
 [   0.       0.       0.963 -153.72 ]
 [   0.       0.       0.       1.   ]]

1.4. Metadata Loss

Converting from DICOM to NIfTI typically drops most clinical metadata (patient, study, device, acquisition parameters). The file becomes detached from its clinical context and is hard to round‑trip safely.

1.5. Limited Compression Efficiency

.nii.gz uses gzip (DEFLATE), a general‑purpose codec not optimized for medical images. Compression ratios are typically 2–4×, leaving room for improvement.

2. DiCube: Designed For Clinical + Research Together

2.1. Unified Coordinate System (LPS+)

DiCube adopts DICOM’s LPS+ consistently. Reading the same DICOM series, DiCube matches SimpleITK’s interpretation.

import os
import pydicom
import dicube, numpy as np
import SimpleITK as sitk

dicom_dir = "dicube-testdata/dicom/sample_200"

# SimpleITK read DICOM series (LPS+)
series_reader = sitk.ImageSeriesReader()
series_ids = series_reader.GetGDCMSeriesIDs(dicom_dir)
series_files = series_reader.GetGDCMSeriesFileNames(dicom_dir, series_ids[0])
series_reader.SetFileNames(series_files)
sitk_image_from_dicom = series_reader.Execute()

# Keep one original DICOM for later round-trip comparisons
original_dcm = pydicom.dcmread(series_files[0], stop_before_pixels=True)

dcb_image = dicube.load_from_dicom_folder(dicom_dir, sort_method=dicube.SortMethod.POSITION_RIGHT_HAND)
print("SimpleITK (from DICOM):", np.round(sitk_image_from_dicom.GetOrigin(), 3))
print("DiCube      (from DICOM):", np.round(dcb_image.space.origin, 3))

SimpleITK (from DICOM): [-102.08  -31.26 1282.2 ]
DiCube      (from DICOM): [-102.08  -31.26 1282.2 ]

2.2. Full Metadata Preservation and Round‑Trip

DiCube embeds all DICOM metadata (including private tags) and supports lossless DICOM round‑trip.

import os
import pydicom

dicube.save(dcb_image, 'dicube-testdata/test.dcbs')
dicube.save_to_dicom_folder(dcb_image, 'dicube-testdata/roundtrip_test')

roundtrip_slice_path = 'dicube-testdata/roundtrip_test/slice_0000.dcm'
roundtrip_dcm = pydicom.dcmread(roundtrip_slice_path)

print("Round‑trip fields match:")
for field in ['PatientID','StudyDate','Manufacturer','WindowWidth','WindowCenter']:
    print(field, original_dcm.get(field) == roundtrip_dcm.get(field))

Round‑trip fields match:
PatientID True
StudyDate True
Manufacturer True
WindowWidth True
WindowCenter True

2.3. Efficient Medical Compression (HTJ2K)

HTJ2K (a DICOM standard codec) provides significantly higher compression ratios and fast throughput compared to gzip for imaging data.

2.4. Optimized I/O

Single‑file structure + efficient codec lead to faster loads than .nii.gz in many scenarios.

import time

start = time.time(); _ = sitk.ReadImage(nifti_file); nifti_ms = (time.time()-start)*1000
start = time.time(); _ = dicube.load('dicube-testdata/test.dcbs'); dcb_ms = (time.time()-start)*1000
print(f"NIfTI load: {nifti_ms:.0f} ms, DiCube load: {dcb_ms:.0f} ms")

NIfTI load: 119 ms, DiCube load: 146 ms

3. Summary

Feature	NIfTI Issue	DiCube Solution
Coordinates	Mixed LPS+/RAS+, qform/sform ambiguity	Unified LPS+ (DICOM‑consistent)
Metadata	>95% clinical metadata lost	100% lossless round‑trip with full tags
Compression	Gzip (2–4×)	HTJ2K (often 5–15× or better)
DICOM Workflow	One‑way lossy conversion	Two‑way lossless; safe clinical container
I/O + API	Gzip‑bound I/O, library variance	Single‑file + clear, consistent APIs

Conclusion: NIfTI works well for pure research exchange. For applications that require integrity, traceability, and clinical interoperability, DiCube offers a safer, faster, and more reliable modern alternative.