Images used by CALMaR

MRI is a family of measurements. Different acquisitions emphasise different tissue properties, so an algorithm's required modality is part of its scientific specification.

Contributors do not need to diagnose scans or recognise every sequence by sight. They do need to know what information an input contributes and whether a tool was designed for it.

Image type is not the analysis

An image type describes how data were acquired or derived—for example, T1-weighted MRI, DWI, an ADC map, or fMRI. An analysis is what is done with those data—for example, segmentation, registration, tractography, connectivity estimation, atlas overlap, or decoding.

The same T1-weighted image could support brain extraction, tissue segmentation, registration, morphometry, lesion segmentation, or atlas mapping. Conversely, one analysis may combine several image types. Never infer the analysis solely from the file's appearance or modality label.

Figure 1. One label, many MRI contrasts. These examples include phase-contrast angiography, susceptibility-weighted imaging, T1-weighted MRI, FLAIR, T2-weighted MRI and perfusion-weighted imaging. They illustrate why “an MRI” is not a sufficiently precise input specification. Reproduced from Figure 1 of MacIntosh and Graham (2013), licensed CC BY 3.0.

CALMaR-relevant image types

Example	Image type	What it helps show	Why the workflow may use it
	T1-weighted MRI	Anatomical structure and tissue boundaries	Chronic-lesion segmentation, brain extraction, registration, atlas mapping
	T2-weighted MRI	Water-sensitive tissue contrast; fluid is usually bright	Complementary lesion information and white-matter abnormality context
	FLAIR MRI	T2-like contrast with free-fluid signal suppressed	Making many lesions and white-matter abnormalities conspicuous near CSF
	Diffusion-weighted imaging	Restricted water diffusion	Detecting acute ischaemic injury
	Apparent diffusion coefficient	Quantified diffusion signal	Interpreting whether DWI hyperintensity reflects true restriction
	Diffusion MRI	Direction-dependent diffusion	Modelling white-matter organisation and structural connectivity
	Functional MRI	Blood-oxygenation changes related to neural activity	Research on task activity and functional networks
	CT	Tissue density and blood	Common acute clinical imaging, requiring modality-specific processing

Table 1. CALMaR-relevant image types. These are real images, included to show that the inputs look materially different—not as a sequence-recognition or diagnostic test. Sources and licences are listed in image credits.

Inputs are not interchangeable

Changing the input modality changes the statistical appearance of tissue and pathology. A model trained on chronic T1-weighted scans has not automatically learned acute DWI, FLAIR, or CT.

Before connecting a tool, record:

• Supported modalities and required combinations
• Stroke type and time window represented in development data
• Expected dimensionality, voxel spacing, orientation, and intensity handling
• Whether the tool returns probabilities, labels, or both
• Preprocessing included in the tool versus required upstream

Multimodal inputs

Multiple images from the same person can offer complementary information, but they must be aligned. A DWI volume, ADC map, FLAIR image, and T1 image may have different grids, distortions, coverage, and spatial resolution.

The workflow must know which image defines the reference space and must preserve every transform used to align the others.

What image brightness does not mean

MRI intensity is generally not a universal physical scale. Brightness can vary with acquisition settings, reconstruction, scanner, coil, preprocessing, and display window. Code should not assume that the same numeric intensity has the same biological meaning across arbitrary scans unless a method explicitly establishes that relationship.

Practical rule

Treat the modality, acquisition context, time since stroke, and spatial metadata as part of the input. The voxel array alone is incomplete.