nifreeze.estimator module

Orchestrates model and registration in volume-to-volume artifact estimation.

class nifreeze.estimator.Estimator(self, model: 'BaseModel | str', strategy: 'str' = 'random', prev: 'Estimator | Filter | None' = None, model_kwargs: 'dict | None' = None, single_fit: 'bool' = False, start_index: 'int' = 0, stop_index: 'int | None' = None, **kwargs)[source]

Bases: object

Orchestrates components for a single estimation step.

The estimator traverses the 4D sequence with a Leave-One-Volume-Out (LOVO) strategy: for each target volume it fits the model on the remaining volumes, predicts the held-out volume, and registers the observed volume to that prediction. Held-out independence is what makes the estimated transform valid (see fit_predict()).

Parameters:

  • model (BaseModel or str) – The model (or model name) used to generate the registration target.

  • strategy (str, optional) – Name of the iterator used to traverse the 4D sequence.

  • prev (Estimator, Filter, optional) – A previous estimator/filter whose output initializes this one (cascade).

  • model_kwargs (dict, optional) – Extra keyword arguments passed through to the model at construction.

  • single_fit (bool, optional) – Run the model in single-fit mode: fit once on all volumes up front (via model.fit_predict(None)) and lock that fit instead of refitting per held-out volume. Prediction is no longer unbiased.

  • start_index (int, optional) – First volume index to process.

  • stop_index (int, optional) – One-past-last volume index to process (None = to the end).

run(dataset: DatasetT, **kwargs) → Self[source]

Trigger execution of the workflow this estimator belongs.

Parameters:

dataset (BaseDataset) – The input dataset this estimator operates on.

Returns:

The estimator, after fitting.

Return type:

Estimator

class nifreeze.estimator.Filter(self, /, *args, **kwargs)[source]

Bases: object

Alters an input data object (e.g., downsampling).

run(dataset: DatasetT, **kwargs) → DatasetT[source]

Trigger execution of the designated filter.

Parameters:

dataset (BaseDataset) – The input dataset this estimator operates on.

Returns:

dataset – The dataset, after filtering.

Return type:

BaseDataset