pyanno4rt.learning_model.features

Features module.

The module aims to provide methods and classes to handle the features of the base data set, i.e., mapping features to segments and definitions from the feature catalogue and iteratively (re)calculate the values as input to the learning model. In addition, the module contains the feature catalogue.

Subpackages

pyanno4rt.learning_model.features.catalogue

Overview

Classes
`FeatureMapGenerator`	Feature map generation class.
`FeatureCalculator`	Feature value and gradient (re)calculation class.

Classes

class pyanno4rt.learning_model.features.FeatureMapGenerator(model_label, fuzzy_matching)[source]

Feature map generation class.

This class provides a mapping between the features from the data set, the structures from the segmentation, and the definitions from the feature catalogue. Matching is based on fuzzy or exact string matching.

Parameters:: fuzzy_matching (bool) – Indicator for the use of fuzzy string matching (if ‘False’, exact string matching is applied).

fuzzy_matching

See ‘Parameters’.

Type:: bool

feature_map

Dictionary with information on the mapping of features in the dataset with the segmented structures and their computation/differentiation functions.

Type:: dict

Notes

In the current implementation, string matching works best if:

names from segments in the segmentation do not have any special characters except “_” (which will automatically be removed before matching);
feature names follow the scheme <name of the segment> _<name of the feature in the catalogue>_<optional parameters>, e.g. “parotidLeft_doseMean” (mean dose to the left parotid) or “parotidRight_doseGradient_x” (dose gradient in x-direction for the right parotid).

Overview

Methods
`generate`(data_information)	Generate the feature map by fuzzy or exact string matching.

Members

generate(data_information)[source]

Generate the feature map by fuzzy or exact string matching.

Parameters:: ...
Returns:: feature_map – Dictionary with information on the mapping of features in the dataset with the segmented structures and their computation/differentiation functions.
Return type:: dict

class pyanno4rt.learning_model.features.FeatureCalculator(write_features, verbose=True)[source]

Feature value and gradient (re)calculation class.

Parameters:: write_features (bool) – Indicator for tracking the feature values.

write_features

See ‘Parameters’.

Type:: bool

feature_history

Feature values per iteration. If write_features is False, this attribute is not set.

Type:: ndarray or None

gradient_history

Gradient matrices per iteration. If write_gradients is False, this attribute is not set.

Type:: list or None

radiomics

Dictionary for mapping the radiomic feature names to the radiomic feature values. It allows to retrieve the feature values after first computation and thus prevents unnecessary recalculation.

Type:: dict

demographics

Dictionary for mapping the demographic feature names to the demographic feature values. It allows to retrieve the feature values after first computation and thus prevents unnecessary recalculation.

Type:: dict

feature_inputs

Dictionary for collecting the candidate feature input values. This allows to centralize the input retrieval for all calculations.

Type:: dict

__iteration__

Iteration numbers for the feature calculation and the optimization problem. By keeping the two elements the same, it is assured that the feature calculator is only active for new problem iterations, rather than per evaluation step.

Type:: list

__dose_cache__

Cache array for the dose values.

Type:: ndarray

__feature_cache__

Cache array for the feature values.

Type:: ndarray

Overview

Methods
`add_feature_map`(feature_map, return_self)	Add the feature map to the calculator.
`precompute`(dose, segment)	Precompute the dose, dose cube and segment masks as inputs for the feature calculation.
`featurize`(dose, segment, no_cache)	Convert dose and segment information into the feature vector.
`get_feature_vector`()	Get the feature vector.
`gradientize`(dose, segment)	Convert dose and segment information into the gradient matrix.

Members

add_feature_map(feature_map, return_self=False)[source]

Add the feature map to the calculator.

Parameters:

feature_map (dict) –

…

precompute(dose, segment)[source]

Precompute the dose, dose cube and segment masks as inputs for the feature calculation.

Parameters:

dose (tuple of ndarray) – Value of the dose for a single or multiple segments.
segment (list of strings) – Names of the segments associated with the dose.

featurize(dose, segment, no_cache=False)[source]

Convert dose and segment information into the feature vector.

Parameters:

dose (tuple of ndarray) – Value of the dose for a single or multiple segments.
segment (list of strings) – Names of the segments associated with the dose.

Returns:

Values of the calculated features.

Return type:

ndarray

get_feature_vector()[source]

Get the feature vector.

Parameters:

dose (tuple of ndarray) – Value of the dose for a single or multiple segments.
segment (list of strings) – Names of the segments associated with the dose.

Returns:

Values of the calculated features.

Return type:

ndarray

gradientize(dose, segment)[source]

Convert dose and segment information into the gradient matrix.

Parameters:

dose (tuple of ndarray) – Value of the dose for a single or multiple segments.
segment (list of strings) – Names of the segments associated with the dose.

Returns:

Matrix of the calculated gradients.

Return type:

csr_matrix