Methods and apparatus for fast identification of relevant features for classification or regression
AssigneePerkinElmer Cellular Technologies Germany GmbH
InventorPeet Kask
Life of this patent
Estimating the term…
Abstract
In various embodiments, methods and apparatus are provided for automated selection of features of cells useful for classifying cell phenotype. The methods include determining a signal-to-noise ratio (S/N) for each of a plurality of pairs of features, rather than S/N for individual features. The approach is capable of quickly identifying a small set of features of imaged cells that are most relevant for classification of a desired cell phenotype from among a very large number of features. The small group of relevant features can then be used to more efficiently and more accurately classify phenotype of unidentified cells.
Claims33
An apparatus for automated selection of features of cells useful for classifying cell phenotype, said features extracted from one or more images of the cells, the apparatus comprising:
- (a) a memory for storing a code defining a set of instructions;
- and
- (b) a processor for executing the set of instructions,
- wherein the code comprises an automated feature selection module to identify a set S2 of N2 calculated features for classification of cell phenotype, wherein the automated image feature selection module is configured to:
- (i) determine a signal-to-noise ratio for each of a plurality of pairs of features from an original set S0 of N0 calculated features of cells in one or more images, wherein the signal-to-noise ratio of a given pair of features reflects an ability to use that pair of features to classify cell phenotype, and identify a predetermined number NP of feature pairs from the original set S0 of N0 calculated features, said NP pairs having the highest signal-to-noise ratio, wherein NP is an integer≧2;
- and
- (ii) identify the set S2 of N2 calculated features for classification of cell phenotype from among the members of the Np pairs of calculated features.
A method for automated selection of image features useful for classifying cell phenotype, the method comprising the steps of:
- (a) quantifying each of a plurality of features for a first set of imaged cells obtained from a subject and having a first known class;
- (b) quantifying each of the plurality of features for a second set of imaged cells obtained from a subject and having a second known class different from the first known class;
- (c) using values of the features quantified in steps (a) and (b) for the first and second sets of imaged cells, determining for each of a plurality of selected pairs of features a signal-to-noise ratio for discrimination between the first and second classes on the basis of the selected pair of features;
- (d) identifying a predetermined number Np of feature pairs with highest signal-to-noise ratio determined in step (c) wherein Np is an integer≧2;
- and
- (e) identifying a set S1 of all features within the Np feature pairs identified in step (d) from which the set S2 of image features for classification of cell phenotype of an unknown sample is selected.
Dependencies are inferred from the claim text.