Skip to content

Harmonization Options and Procedures

Brunno M de Campos edited this page Sep 10, 2026 · 3 revisions

S²M provides several options for image harmonization, always performed relative to the currently loaded Reference Dataset. Some of these options were specifically designed for situations in which FLAIR images are included in the Focal Cortical Dysplasia modality, whereas others provide more general approaches that can be adapted to different datasets and acquisition conditions.

image

The available options are:

  1. Estimate using loaded cases

Harmonization parameters are estimated directly from the currently loaded cases. This approach uses the available sample to model and correct for shared biases, such as those related to the scanner, acquisition protocol, or FLAIR-related effects.

While convenient, this method should be used with caution. Effects that are common across the sample, including potential pathological patterns, may influence the parameter estimation and could therefore be partially attenuated.

In this sense, a good practice is to include control subjects in the sample used for parameter estimation. This can be done in two different ways:

a) Separate S²M session using only control subjects

A separate S²M session can be performed using only control subjects to estimate harmonization parameters. In this case, the estimated parameters will primarily characterize image- or center-related biases and can subsequently be applied to the study sample using Option 2: Add previously estimated parameters.

b) Controls and patients loaded together

Alternatively, controls and patients can be loaded together in the same S²M session. S²M will ask how the loaded subjects should be used for parameter estimation.

image

image

Each subject can be either included in the parameter estimation (e.g., the control subjects within the loaded sample) or excluded from the estimation using a tabulated file with a binary column vector (xls, txt, csv...). Regardless of which subjects are used to estimate the parameters, the resulting harmonization correction will be applied to the entire loaded sample.

For both approaches described above, S²M enables harmonization parameter estimation only when 10 or more cases are loaded. To harmonize a single loaded case, harmonization parameters must have been estimated previously, ideally from a sample acquired using equivalent imaging conditions and protocols.

  1. Add previously estimated parameters

Harmonization is performed using parameters previously estimated from an external dataset. Ideally, these parameters should be derived from images acquired under similar conditions, including the same scanner, protocol, and sequence characteristics.

This is the recommended approach when suitable control data are available, as it reduces potential bias while minimizing the risk of attenuating relevant subject-specific alterations.

  1. Use S²M native FLAIR bias parameters

S²M provides predefined harmonization parameters specifically designed to correct for FLAIR-related biases. These parameters were estimated by comparing the S²M Reference Dataset with a control group acquired using similar imaging protocols and exhibiting similar FLAIR-related bias.

This option is intended as a fallback when user-defined harmonization is not feasible. It helps mitigate FLAIR-induced biases while preserving the potential advantages of including FLAIR information.

This option is available only in the Focal Cortical Dysplasia Modality and only when FLAIR images have been added.

Clone this wiki locally